Initial commit

backend/.python-version

@@ -0,0 +1 @@
+3.12

backend/Dockerfile

@@ -0,0 +1,20 @@
+# Simple Dockerfile - install everything in one place
+FROM python:3.12-slim-bookworm
+
+WORKDIR /app
+
+ENV PYTHONUNBUFFERED=1 \
+    PYTHONDONTWRITEBYTECODE=1
+
+# Install pip packages directly (no uv complexity)
+COPY pyproject.toml ./
+RUN pip install --no-cache-dir -e .
+
+# Copy application
+COPY . .
+
+# Expose port
+EXPOSE 8000
+
+# Run directly
+CMD ["python", "-m", "uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8000"]

backend/app/api/v1/analysis.py

@@ -0,0 +1,147 @@
+from fastapi import APIRouter, HTTPException
+from pydantic import BaseModel
+from typing import List, Any, Dict, Optional
+import pandas as pd
+import numpy as np
+from app.core.engine.clean import detect_univariate_outliers, detect_multivariate_outliers, merge_outliers, merge_outliers_structured
+from app.core.engine.stats import calculate_correlation_matrix, calculate_feature_importance, run_regression_analysis
+
+router = APIRouter(prefix="/analysis", tags=["analysis"])
+
+class TypeValidationRequest(BaseModel):
+    data: List[Any]
+    target_type: str
+
+class OutlierDetectionRequest(BaseModel):
+    data: List[Dict[str, Optional[Any]]]
+    columns: List[str]
+    method: str = "both"
+    excluded_indices: List[int] = []  # Rows to exclude from outlier detection
+
+class CorrelationRequest(BaseModel):
+    data: List[Dict[str, Optional[Any]]]
+    columns: List[str]
+    method: str = "pearson"  # pearson, spearman, kendall
+    min_threshold: Optional[float] = None  # Optional minimum correlation threshold
+    include_pvalues: bool = True
+
+class FeatureImportanceRequest(BaseModel):
+    data: List[Dict[str, Optional[Any]]]
+    features: List[str]
+    target: str
+
+class RegressionRequest(BaseModel):
+    data: List[Dict[str, Optional[Any]]]
+    x_features: List[str]
+    y_target: str
+    model_type: str = "linear"
+    # New Engineering Parameters
+    poly_degree: int = 1 # Default to linear
+    include_interactions: bool = False
+
+@router.post("/validate-type")
+async def validate_type_conversion(request: TypeValidationRequest):
+    s = pd.Series(request.data)
+    try:
+        if request.target_type == "numeric":
+            pd.to_numeric(s, errors='raise')
+        elif request.target_type == "date":
+            pd.to_datetime(s, errors='raise')
+        return {"status": "ok", "valid": True}
+    except Exception as e:
+        return {"status": "error", "valid": False, "message": str(e)}
+
+@router.post("/detect-outliers")
+async def detect_outliers(request: OutlierDetectionRequest):
+    if not request.data:
+        return {"outliers": []}
+
+    df = pd.DataFrame(request.data).fillna(np.nan)
+
+    # Pass excluded indices to detection functions
+    uni_results = detect_univariate_outliers(
+        df, request.columns, request.excluded_indices
+    ) if request.method in ["univariate", "both"] else {}
+
+    multi_results = detect_multivariate_outliers(
+        df, request.columns, request.excluded_indices
+    ) if request.method in ["multivariate", "both"] else {}
+
+    # Use the new structured merge function
+    structured = merge_outliers_structured(uni_results, multi_results)
+
+    return {
+        "status": "ok",
+        "total_count": len(structured["all"]),
+        "outliers": structured["all"],  # Backwards compatibility
+        "univariate": structured["univariate"],  # New: Column-specific outliers
+        "multivariate": structured["multivariate"]  # New: Global outliers
+    }
+
+@router.post("/correlation")
+async def get_correlation(request: CorrelationRequest):
+    if not request.data or not request.columns:
+        return {
+            "status": "error",
+            "message": "Data and columns are required",
+            "result": {"matrix": [], "pvalues": [], "metadata": {}}
+        }
+
+    df = pd.DataFrame(request.data).fillna(np.nan)
+
+    # Validate method parameter
+    valid_methods = ['pearson', 'spearman', 'kendall']
+    if request.method not in valid_methods:
+        raise HTTPException(
+            status_code=400,
+            detail=f"Invalid method. Choose from: {', '.join(valid_methods)}"
+        )
+
+    try:
+        result = calculate_correlation_matrix(
+            df,
+            request.columns,
+            method=request.method,
+            min_threshold=request.min_threshold,
+            include_pvalues=request.include_pvalues
+        )
+
+        # Add summary statistics
+        from app.core.engine.stats import get_correlation_summary
+        summary = get_correlation_summary(result)
+
+        return {
+            "status": "ok",
+            "result": result,
+            "summary": summary
+        }
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Correlation calculation failed: {str(e)}")
+
+@router.post("/feature-importance")
+async def get_feature_importance(request: FeatureImportanceRequest):
+    if not request.data or not request.features or not request.target: return {"importances": []}
+    df = pd.DataFrame(request.data).fillna(np.nan)
+    return {"status": "ok", "importances": calculate_feature_importance(df, request.features, request.target)}
+
+@router.post("/run-regression")
+async def run_regression(request: RegressionRequest):
+    if not request.data or not request.x_features or not request.y_target:
+        raise HTTPException(status_code=400, detail="Incomplete parameters.")
+    
+    df = pd.DataFrame(request.data).fillna(np.nan)
+    
+    try:
+        results = run_regression_analysis(
+            df, 
+            request.x_features, 
+            request.y_target, 
+            request.model_type,
+            request.poly_degree,
+            request.include_interactions
+        )
+        return {"status": "ok", "results": results}
+    except ValueError as e:
+        raise HTTPException(status_code=400, detail=str(e))
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Internal Analysis Error: {str(e)}")

backend/app/api/v1/reports.py

@@ -0,0 +1,33 @@
+from fastapi import APIRouter, HTTPException, Response
+from pydantic import BaseModel
+from typing import Dict, Any, List
+from app.core.engine.reports import create_pdf_report
+
+router = APIRouter(prefix="/reports", tags=["reporting"])
+
+class ExportRequest(BaseModel):
+    project_name: str
+    results: Dict[str, Any]
+    audit_trail: Dict[str, Any]
+
+@router.post("/export")
+async def export_report(request: ExportRequest):
+    """
+    Generates and returns a PDF report.
+    """
+    try:
+        pdf_bytes = create_pdf_report(
+            request.project_name, 
+            request.results, 
+            request.audit_trail
+        )
+        
+        return Response(
+            content=pdf_bytes,
+            media_type="application/pdf",
+            headers={
+                "Content-Disposition": f"attachment; filename=Report_{request.project_name}.pdf"
+            }
+        )
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))

backend/app/api/v1/upload.py

@@ -0,0 +1,44 @@
+from fastapi import APIRouter, UploadFile, File, HTTPException
+from fastapi.responses import StreamingResponse
+import io
+import json
+from app.core.engine.ingest import parse_file, get_column_metadata, dataframe_to_arrow_stream
+
+router = APIRouter(prefix="/upload", tags=["ingestion"])
+
+@router.post("")
+async def upload_file(file: UploadFile = File(...)):
+    """
+    Endpoint to upload Excel/CSV files and receive an Apache Arrow stream.
+    Metadata about columns is sent in the X-Column-Metadata header.
+    """
+    # 1. Validation
+    if not file.filename.endswith(('.xlsx', '.xls', '.csv')):
+        raise HTTPException(status_code=400, detail="Only .xlsx, .xls and .csv files are supported.")
+    
+    try:
+        content = await file.read()
+        
+        # 2. Parsing
+        df = parse_file(content, file.filename)
+        
+        # 3. Metadata Extraction
+        metadata = get_column_metadata(df)
+        
+        # 4. Conversion to Arrow
+        arrow_bytes = dataframe_to_arrow_stream(df)
+        
+        # We use a StreamingResponse to send the binary Arrow data.
+        # Metadata is sent as a custom header (JSON stringified).
+        return StreamingResponse(
+            io.BytesIO(arrow_bytes),
+            media_type="application/vnd.apache.arrow.stream",
+            headers={
+                "X-Column-Metadata": json.dumps(metadata),
+                "Access-Control-Expose-Headers": "X-Column-Metadata"
+            }
+        )
+        
+    except Exception as e:
+        # In a real app, we'd log this properly
+        raise HTTPException(status_code=400, detail=f"Error processing file: {str(e)}")

backend/app/core/engine/clean.py

@@ -0,0 +1,165 @@
+import pandas as pd
+import numpy as np
+from sklearn.ensemble import IsolationForest
+from typing import List, Dict, Any
+
+def detect_univariate_outliers(df: pd.DataFrame, columns: List[str], excluded_indices: List[int] = None) -> Dict[int, List[str]]:
+    """
+    Detects outliers in specific numeric columns using the Interquartile Range (IQR) method.
+
+    Args:
+        df: Input DataFrame
+        columns: List of column names to analyze
+        excluded_indices: List of row indices to exclude from detection
+
+    Returns:
+        Dictionary of {original_row_index: [reasons]}
+    """
+    # Exclude specified rows if provided
+    if excluded_indices:
+        df = df[~df.index.isin(excluded_indices)]
+
+    outliers = {}
+    for col in columns:
+        if col not in df.columns:
+            continue
+
+        s = pd.to_numeric(df[col], errors='coerce')
+        q1 = s.quantile(0.25)
+        q3 = s.quantile(0.75)
+        iqr = q3 - q1
+        lower_bound = q1 - 1.5 * iqr
+        upper_bound = q3 + 1.5 * iqr
+
+        flags = (s < lower_bound) | (s > upper_bound)
+        indices = df.index[flags].tolist()
+
+        for idx in indices:
+            val = df.at[idx, col]
+            reason = f"Column '{col}' value {val} is outside IQR bounds [{lower_bound:.2f}, {upper_bound:.2f}]"
+            if idx not in outliers:
+                outliers[idx] = []
+            outliers[idx].append(reason)
+
+    return outliers
+
+def detect_multivariate_outliers(df: pd.DataFrame, columns: List[str], excluded_indices: List[int] = None) -> Dict[int, List[str]]:
+    """
+    Detects anomalies across multiple numeric columns using Isolation Forest.
+
+    Args:
+        df: Input DataFrame
+        columns: List of column names to analyze
+        excluded_indices: List of row indices to exclude from detection (already filtered out)
+
+    Returns:
+        Dictionary of {original_row_index: [reasons]}
+    """
+    # Store original indices
+    original_indices = df.index.tolist()
+
+    # Exclude specified rows if provided
+    if excluded_indices:
+        df = df[~df.index.isin(excluded_indices)]
+
+    # Select only relevant numeric columns
+    numeric_df = df[columns].apply(pd.to_numeric, errors='coerce')
+
+    if numeric_df.empty:
+        return {}
+
+    # Get rows with all values present (no NaNs in selected columns)
+    valid_mask = numeric_df.notna().all(axis=1)
+    numeric_df_clean = numeric_df[valid_mask]
+
+    if numeric_df_clean.empty:
+        return {}
+
+    # Fit Isolation Forest
+    model = IsolationForest(contamination='auto', random_state=42)
+    preds = model.fit_predict(numeric_df_clean)
+
+    # IsolationForest returns -1 for outliers
+    # Get the indices from the clean DataFrame (these are the original indices)
+    outlier_indices = numeric_df_clean.index[preds == -1].tolist()
+
+    return {int(idx): ["Multivariate anomaly detected by Isolation Forest"] for idx in outlier_indices}
+
+def merge_outliers(uni: Dict[int, List[str]], multi: Dict[int, List[str]]) -> List[Dict[str, Any]]:
+    """
+    Merges results into a flat list of outlier objects.
+    DEPRECATED: Use merge_outliers_structured instead for better type separation.
+    """
+    all_indices = set(uni.keys()) | set(multi.keys())
+    results = []
+
+    for idx in sorted(all_indices):
+        reasons = uni.get(idx, []) + multi.get(idx, [])
+        results.append({
+            "index": int(idx),
+            "reasons": reasons
+        })
+
+    return results
+
+def merge_outliers_structured(uni: Dict[int, List[str]], multi: Dict[int, List[str]]) -> Dict[str, Any]:
+    """
+    Merges and separates outliers by type for better frontend handling.
+
+    Returns:
+        Dictionary with:
+        - 'univariate': Dict mapping column names to their specific outliers
+        - 'multivariate': List of outliers that affect multiple columns
+        - 'all': Flat list of all outliers (for backwards compatibility)
+
+    This structure allows the frontend to:
+    1. Show column-specific outliers when clicking a column header
+    2. Show global/multivariate outliers in a separate view
+    3. Clearly distinguish between local and global anomalies
+    """
+    # Extract column names from univariate reasons
+    column_outliers: Dict[str, List[Dict[str, Any]]] = {}
+
+    for idx, reasons in uni.items():
+        for reason in reasons:
+            # Extract column name from reason string
+            # Format: "Column 'Price' value 100 is outside..."
+            if "Column '" in reason:
+                col_start = reason.index("Column '") + 8
+                col_end = reason.index("'", col_start)
+                col_name = reason[col_start:col_end]
+
+                if col_name not in column_outliers:
+                    column_outliers[col_name] = []
+
+                # Check if this row index already exists for this column
+                existing = next((x for x in column_outliers[col_name] if x["index"] == idx), None)
+                if not existing:
+                    column_outliers[col_name].append({
+                        "index": int(idx),
+                        "reasons": [reason]
+                    })
+                else:
+                    existing["reasons"].append(reason)
+
+    # Prepare multivariate outliers
+    multivariate_list = [
+        {"index": int(idx), "reasons": reasons}
+        for idx, reasons in multi.items()
+    ]
+
+    # Prepare legacy flat format (backwards compatibility)
+    all_indices = set(uni.keys()) | set(multi.keys())
+    all_outliers = []
+    for idx in sorted(all_indices):
+        reasons = uni.get(idx, []) + multi.get(idx, [])
+        all_outliers.append({
+            "index": int(idx),
+            "reasons": reasons
+        })
+
+    return {
+        "univariate": column_outliers,
+        "multivariate": multivariate_list,
+        "all": all_outliers
+    }

backend/app/core/engine/ingest.py

@@ -0,0 +1,56 @@
+import pandas as pd
+import pyarrow as pa
+import io
+from typing import Tuple, Dict, Any
+
+def parse_file(file_content: bytes, filename: str) -> pd.DataFrame:
+    """
+    Parses the uploaded file (Excel or CSV) into a Pandas DataFrame.
+    """
+    file_obj = io.BytesIO(file_content)
+    
+    if filename.endswith(('.xlsx', '.xls')):
+        df = pd.read_excel(file_obj)
+    elif filename.endswith('.csv'):
+        # Attempt to detect common delimiters if needed, default to comma
+        df = pd.read_csv(file_obj)
+    else:
+        raise ValueError(f"Unsupported file format: {filename}")
+    
+    # Basic hygiene: strip whitespace from headers
+    df.columns = [str(c).strip() for c in df.columns]
+    
+    return df
+
+def get_column_metadata(df: pd.DataFrame) -> list:
+    """
+    Returns a list of column metadata (name and inferred type).
+    """
+    metadata = []
+    for col in df.columns:
+        dtype = str(df[col].dtype)
+        # Simplify types for the frontend
+        inferred_type = "numeric"
+        if "object" in dtype or "string" in dtype:
+            inferred_type = "categorical"
+        elif "datetime" in dtype:
+            inferred_type = "date"
+        elif "bool" in dtype:
+            inferred_type = "boolean"
+            
+        metadata.append({
+            "name": col,
+            "type": inferred_type,
+            "native_type": dtype
+        })
+    return metadata
+
+def dataframe_to_arrow_stream(df: pd.DataFrame) -> bytes:
+    """
+    Converts a Pandas DataFrame to an Apache Arrow IPC stream.
+    """
+    table = pa.Table.from_pandas(df)
+    sink = pa.BufferOutputStream()
+    with pa.ipc.new_stream(sink, table.schema) as writer:
+        writer.write_table(table)
+    return sink.getvalue().to_pybytes()

backend/app/core/engine/reports.py

@@ -0,0 +1,223 @@
+from fpdf import FPDF
+from datetime import datetime
+from io import BytesIO
+import pandas as pd
+import platform
+import sklearn
+import statsmodels
+import os
+import matplotlib
+matplotlib.use('Agg')  # Use non-GUI backend
+import matplotlib.pyplot as plt
+import tempfile
+
+class AnalysisReport(FPDF):
+    def header(self):
+        try:
+            # Add Unicode font support for accented characters
+            self.set_font('Arial', 'B', 15)
+            self.set_text_color(79, 70, 229) # Indigo 600
+            self.cell(0, 10, 'Data_analysis - Rapport de Validation', 0, 1, 'L')
+            self.set_draw_color(226, 232, 240)
+            self.line(10, 22, 200, 22)
+            self.ln(10)
+        except Exception as e:
+            print(f"Header error: {e}")
+
+    def footer(self):
+        try:
+            self.set_y(-15)
+            self.set_font('Arial', 'I', 8)
+            self.set_text_color(148, 163, 184)
+            self.cell(0, 10, f'Page {self.page_no()} | Genere le {datetime.now().strftime("%Y-%m-%d %H:%M")}', 0, 0, 'C')
+        except Exception as e:
+            print(f"Footer error: {e}")
+
+def create_pdf_report(project_name: str, results: dict, audit_trail: dict) -> bytes:
+    try:
+        pdf = AnalysisReport()
+        pdf.add_page()
+
+        # 1. Summary
+        pdf.set_font('Arial', 'B', 12)
+        pdf.set_text_color(51, 65, 85)
+        pdf.cell(0, 10, f"Projet : {project_name}", 0, 1)
+        pdf.ln(5)
+
+        # Handle missing model_type
+        model_type = results.get('model_type', 'Regression')
+        if isinstance(model_type, list):
+            model_type = model_type[0] if model_type else 'Regression'
+
+        pdf.set_font('Arial', '', 10)
+        pdf.cell(0, 8, f"Modele : {model_type}", 0, 1)
+
+        # Handle r_squared safely
+        r_squared = results.get('r_squared', 0)
+        if r_squared is None:
+            r_squared = 0
+        pdf.cell(0, 8, f"Precision (R²) : {float(r_squared):.4f}", 0, 1)
+
+        # Handle sample_size safely
+        sample_size = results.get('sample_size', 0)
+        if sample_size is None:
+            sample_size = 0
+        pdf.cell(0, 8, f"Taille de l'echantillon : {int(sample_size)}", 0, 1)
+
+        pdf.ln(10)
+
+        # 2. Coefficients Table
+        pdf.set_font('Arial', 'B', 11)
+        pdf.cell(0, 10, "Coefficients du Modele", 0, 1)
+
+        pdf.set_font('Arial', 'B', 9)
+        pdf.set_fill_color(248, 250, 252)
+        pdf.cell(80, 8, "Feature", 1, 0, 'L', True)
+        pdf.cell(50, 8, "Coefficient", 1, 0, 'R', True)
+        pdf.cell(50, 8, "P-Value", 1, 1, 'R', True)
+
+        # Get coefficients and p_values safely
+        coefficients = results.get('coefficients', {})
+        p_values = results.get('p_values', {})
+
+        if coefficients:
+            pdf.set_font('Arial', '', 9)
+            for name, coef in coefficients.items():
+                # Convert coef to float safely
+                try:
+                    coef_val = float(coef)
+                except (TypeError, ValueError):
+                    coef_val = 0.0
+
+                # Get p-value safely
+                p_val = p_values.get(name, 1.0)
+                try:
+                    p_val = float(p_val)
+                except (TypeError, ValueError):
+                    p_val = 1.0
+
+                pdf.cell(80, 8, str(name), 1)
+                pdf.cell(50, 8, f"{coef_val:.4f}", 1, 0, 'R')
+
+                if p_val < 0.05:
+                    pdf.set_text_color(16, 185, 129) # Emerald
+                else:
+                    pdf.set_text_color(244, 63, 94) # Rose
+                pdf.cell(50, 8, f"{p_val:.4f}", 1, 1, 'R')
+                pdf.set_text_color(51, 65, 85)
+        else:
+            pdf.set_font('Arial', '', 9)
+            pdf.cell(0, 8, "Aucun coefficient disponible", 0, 1)
+
+        pdf.ln(15)
+
+        # 3. Visualization Charts
+        if 'fit_plot' in results and len(results['fit_plot']) > 0:
+            pdf.set_font('Arial', 'B', 11)
+            pdf.cell(0, 10, "Courbe de Regression", 0, 1)
+            pdf.ln(5)
+
+            # Create fit plot
+            fit_data = results['fit_plot']
+            x_vals = [p['x'] for p in fit_data]
+            y_real = [p['real'] for p in fit_data]
+            y_pred = [p['pred'] for p in fit_data]
+
+            plt.figure(figsize=(10, 6))
+            plt.scatter(x_vals, y_real, alpha=0.6, color='#4f46e5', label='Données réelles', s=50)
+            plt.plot(x_vals, y_pred, color='#ef4444', linewidth=2, label='Courbe de régression')
+            plt.xlabel('Valeur X', fontsize=12)
+            plt.ylabel('Valeur Y', fontsize=12)
+            plt.title('Ajustement du Modèle', fontsize=14, fontweight='bold')
+            plt.legend()
+            plt.grid(True, alpha=0.3)
+            plt.tight_layout()
+
+            # Save plot to temp file and add to PDF
+            with tempfile.NamedTemporaryFile(suffix='.png', delete=False) as tmp:
+                plt.savefig(tmp.name, dpi=150, bbox_inches='tight')
+                plt.close()
+                pdf.image(tmp.name, x=10, w=190)
+                os.unlink(tmp.name)
+
+            pdf.ln(10)
+
+        # Residuals plot
+        if 'diagnostic_plot' in results and len(results['diagnostic_plot']) > 0:
+            pdf.set_font('Arial', 'B', 11)
+            pdf.cell(0, 10, "Graphique des Residus", 0, 1)
+            pdf.ln(5)
+
+            residuals_data = results['diagnostic_plot']
+            fitted = [p['fitted'] for p in residuals_data]
+            residuals = [p['residual'] for p in residuals_data]
+
+            plt.figure(figsize=(10, 6))
+            plt.scatter(fitted, residuals, alpha=0.6, color='#4f46e5', s=50)
+            plt.axhline(y=0, color='#ef4444', linestyle='--', linewidth=2)
+            plt.xlabel('Valeurs Ajustees', fontsize=12)
+            plt.ylabel('Residus', fontsize=12)
+            plt.title('Graphique des Residus', fontsize=14, fontweight='bold')
+            plt.grid(True, alpha=0.3)
+            plt.tight_layout()
+
+            with tempfile.NamedTemporaryFile(suffix='.png', delete=False) as tmp:
+                plt.savefig(tmp.name, dpi=150, bbox_inches='tight')
+                plt.close()
+                pdf.image(tmp.name, x=10, w=190)
+                os.unlink(tmp.name)
+
+            pdf.ln(10)
+
+        # 4. Audit Trail (Reproducibility)
+        pdf.set_font('Arial', 'B', 11)
+        pdf.cell(0, 10, "Piste d'Audit & Reproductibilite", 0, 1)
+
+        pdf.set_font('Arial', '', 8)
+        pdf.set_text_color(100, 116, 139)
+
+        # Cleaning steps
+        excluded_count = audit_trail.get('excluded_rows_count', 0)
+        if excluded_count is None:
+            excluded_count = 0
+        pdf.multi_cell(0, 6, f"- Nettoyage : {int(excluded_count)} lignes exclues de l'analyse.")
+
+        # Environment
+        pdf.ln(5)
+        pdf.set_font('Arial', 'B', 8)
+        pdf.cell(0, 6, "Environnement Technique :", 0, 1)
+        pdf.set_font('Arial', '', 8)
+        pdf.cell(0, 5, f"- Python : {platform.python_version()}", 0, 1)
+        pdf.cell(0, 5, f"- Pandas : {pd.__version__}", 0, 1)
+
+        # Try to get sklearn version safely
+        try:
+            pdf.cell(0, 5, f"- Scikit-learn : {sklearn.__version__}", 0, 1)
+        except Exception:
+            pdf.cell(0, 5, "- Scikit-learn : Installé", 0, 1)
+
+        # Try to get statsmodels version safely
+        try:
+            pdf.cell(0, 5, f"- Statsmodels : {statsmodels.__version__}", 0, 1)
+        except Exception:
+            pdf.cell(0, 5, "- Statsmodels : Installé", 0, 1)
+
+        pdf.cell(0, 5, f"- Random Seed : 42 (Fixed)", 0, 1)
+
+        # Generate PDF bytes using BytesIO
+        pdf_buffer = BytesIO()
+        pdf.output(pdf_buffer)
+        return pdf_buffer.getvalue()
+
+    except Exception as e:
+        # Return error as PDF with message using BytesIO
+        error_pdf = FPDF()
+        error_pdf.add_page()
+        error_pdf.set_font('Arial', 'B', 16)
+        error_pdf.cell(0, 10, f"Erreur lors de la generation du PDF", 0, 1)
+        error_pdf.ln(10)
+        error_pdf.set_font('Arial', '', 12)
+        error_pdf.multi_cell(0, 10, f"Erreur: {str(e)}")
+        error_buffer = BytesIO()
+        error_pdf.output(error_buffer)
+        return error_buffer.getvalue()

backend/app/core/engine/stats.py

@@ -0,0 +1,430 @@
+import pandas as pd
+import numpy as np
+import statsmodels.api as sm
+from sklearn.ensemble import RandomForestRegressor
+from sklearn.inspection import permutation_importance
+from sklearn.preprocessing import PolynomialFeatures
+from scipy import stats
+from typing import List, Dict, Any, Tuple
+import sympy as sp
+
+def calculate_correlation_matrix(
+    df: pd.DataFrame,
+    columns: List[str],
+    method: str = 'pearson',
+    min_threshold: float = None,
+    include_pvalues: bool = True
+) -> Dict[str, Any]:
+    """
+    Calculate correlation matrix with optional p-values and filtering.
+
+    Args:
+        df: Input DataFrame
+        columns: List of column names to analyze
+        method: Correlation method ('pearson', 'spearman', 'kendall')
+        min_threshold: Minimum absolute correlation value to include (optional)
+        include_pvalues: Whether to calculate statistical significance
+
+    Returns:
+        Dictionary with matrix data, p-values, and metadata
+    """
+    if not columns:
+        return {"matrix": [], "pvalues": [], "metadata": {}}
+
+    # Convert to numeric and handle missing values
+    numeric_df = df[columns].apply(pd.to_numeric, errors='coerce')
+
+    # Remove columns with too many missing values (>50%)
+    missing_ratios = numeric_df.isnull().sum() / len(numeric_df)
+    valid_cols = missing_ratios[missing_ratios <= 0.5].index.tolist()
+
+    if len(valid_cols) < 2:
+        return {"matrix": [], "pvalues": [], "metadata": {"error": "Need at least 2 valid numeric columns"}}
+
+    # Use pairwise deletion for correlation (more robust than listwise)
+    clean_df = numeric_df[valid_cols]
+
+    # Calculate correlation matrix
+    corr_matrix = clean_df.corr(method=method)
+
+    # Calculate p-values if requested
+    pvalue_matrix = None
+    if include_pvalues:
+        pvalue_matrix = pd.DataFrame(np.zeros_like(corr_matrix),
+                                    index=corr_matrix.index,
+                                    columns=corr_matrix.columns)
+
+        for i, col1 in enumerate(corr_matrix.columns):
+            for j, col2 in enumerate(corr_matrix.index):
+                if i != j:
+                    # Pairwise complete observations
+                    valid_data = clean_df[[col1, col2]].dropna()
+                    if len(valid_data) >= 3:
+                        if method == 'pearson':
+                            _, pval = stats.pearsonr(valid_data.iloc[:, 0], valid_data.iloc[:, 1])
+                        elif method == 'spearman':
+                            _, pval = stats.spearmanr(valid_data.iloc[:, 0], valid_data.iloc[:, 1])
+                        elif method == 'kendall':
+                            _, pval = stats.kendalltau(valid_data.iloc[:, 0], valid_data.iloc[:, 1])
+                        else:
+                            pval = np.nan
+                        pvalue_matrix.iloc[i, j] = pval
+
+    # Build results
+    results = []
+    pvalue_results = []
+
+    for x in corr_matrix.columns:
+        for y in corr_matrix.index:
+            value = float(corr_matrix.at[y, x])
+
+            # Apply threshold filter if specified
+            if min_threshold is not None and abs(value) < min_threshold:
+                continue
+
+            results.append({
+                "x": x,
+                "y": y,
+                "value": value,
+                "abs_value": abs(value)
+            })
+
+            if include_pvalues and pvalue_matrix is not None:
+                pvalue_results.append({
+                    "x": x,
+                    "y": y,
+                    "pvalue": float(pvalue_matrix.at[y, x]) if not pd.isna(pvalue_matrix.at[y, x]) else None,
+                    "significant": bool((pvalue_matrix.at[y, x] or 1) < 0.05 if not pd.isna(pvalue_matrix.at[y, x]) else False)
+                })
+
+    # Calculate summary statistics
+    n_observations = len(clean_df)
+
+    return {
+        "matrix": results,
+        "pvalues": pvalue_results if include_pvalues else [],
+        "metadata": {
+            "method": method,
+            "n_observations": n_observations,
+            "n_variables": len(valid_cols),
+            "columns_analyzed": valid_cols,
+            "threshold_applied": min_threshold
+        }
+    }
+
+def get_correlation_summary(correlation_data: Dict[str, Any]) -> Dict[str, Any]:
+    """
+    Generate summary statistics from correlation data.
+    Identifies strongest correlations (positive and negative).
+    """
+    matrix = correlation_data.get("matrix", [])
+
+    # Filter out diagonal (self-correlation)
+    off_diagonal = [m for m in matrix if m["x"] != m["y"]]
+
+    if not off_diagonal:
+        return {"strongest": [], "weakest": []}
+
+    # Sort by absolute correlation value
+    sorted_by_abs = sorted(off_diagonal, key=lambda x: x["abs_value"], reverse=True)
+
+    # Get strongest correlations (top 5)
+    strongest = sorted_by_abs[:5]
+
+    # Get weakest correlations (bottom 5, but non-zero)
+    weakest = [m for m in sorted_by_abs if m["abs_value"] > 0][-5:]
+    weakest = sorted(weakest, key=lambda x: x["abs_value"])
+
+    return {
+        "strongest": strongest,
+        "weakest": weakest,
+        "total_pairs": len(off_diagonal)
+    }
+
+def calculate_feature_importance(df: pd.DataFrame, features: List[str], target: str) -> List[Dict[str, Any]]:
+    if not features or not target: return []
+    df_clean = df.dropna(subset=[target])
+    X = df_clean[features].apply(pd.to_numeric, errors='coerce').fillna(0)
+    y = df_clean[target]
+    if y.dtype == 'object' or y.dtype == 'string': y = pd.factorize(y)[0]
+    model = RandomForestRegressor(n_estimators=100, random_state=42)
+    model.fit(X, y)
+    result = permutation_importance(model, X, y, n_repeats=10, random_state=42, n_jobs=-1)
+    importances = result.importances_mean
+    results = [{"feature": name, "score": max(0, float(score))} for name, score in zip(features, importances)]
+    total = sum(r["score"] for r in results)
+    if total > 0:
+        for r in results: r["score"] /= total
+    return sorted(results, key=lambda x: x["score"], reverse=True)
+
+def generate_equations(coefficients: Dict[str, float], model_type: str) -> Dict[str, str]:
+    """
+    Generate equation strings in LaTeX, Python, and Excel formats.
+
+    Args:
+        coefficients: Dictionary of feature names to coefficient values
+        model_type: Type of regression model ('linear', 'polynomial', 'exponential', 'logistic')
+
+    Returns:
+        Dictionary with 'latex', 'python', and 'excel' equation strings
+    """
+    from sympy import symbols, sympify, latex, Float, preorder_traversal, Mul, Pow
+
+    # Extract intercept
+    intercept = 0.0
+    feature_coefs = {}
+
+    for key, value in coefficients.items():
+        if key in ['const', 'intercept', '(Intercept)']:
+            intercept = float(value)
+        else:
+            feature_coefs[key] = float(value)
+
+    # Helper function to format number cleanly for Python/Excel
+    def format_number(num: float) -> str:
+        """Format number with 3 decimal places max"""
+        if num == 0:
+            return "0"
+        abs_num = abs(num)
+        # Use scientific notation for very small or very large numbers
+        if abs_num >= 10000 or (abs_num < 0.001 and abs_num > 0):
+            return f"{num:.2e}"
+        # Regular decimal with 3 decimal places max
+        formatted = f"{num:.3f}"
+        # Remove trailing zeros
+        return formatted.rstrip('0').rstrip('.')
+
+    # Build LaTeX with sympy using scientific notation
+    # Create symbols for each variable
+    for name in feature_coefs.keys():
+        safe_name = name.replace(' ', '_').replace('^', '_pow_')
+        symbols(safe_name)
+
+    # Build expression string
+    expr_parts = []
+    intercept_str = f"{intercept:.10f}"
+    expr_parts.append(intercept_str)
+
+    for name, coef in feature_coefs.items():
+        safe_name = name.replace(' ', '_').replace('^', '_pow_')
+        coef_str = f"{coef:.10f}"
+        expr_parts.append(f"{coef_str}*{safe_name}")
+
+    expr_str = " + ".join(expr_parts)
+    expr = sympify(expr_str)
+
+    # Scientific notation rounding function
+    def scientific_round_expr(e, ndigits=2):
+        """
+        Convert floats to scientific notation with specified decimal places.
+        Example: 12345.678 -> 1.23 × 10^4
+        """
+        repl = {}
+        for node in preorder_traversal(e):
+            if isinstance(node, Float):
+                val = float(node.evalf(6))  # Get enough precision
+                abs_val = abs(val)
+
+                # Use scientific notation for large or small numbers
+                if abs_val >= 10000 or (abs_val < 0.01 and abs_val > 0):
+                    sci_str = f"{val:.{ndigits}e}"
+                    mantissa, exponent = sci_str.split('e')
+                    # Reconstruct as: mantissa × 10^exponent
+                    repl[node] = Mul(Float(mantissa), Pow(10, int(exponent)), evaluate=False)
+                else:
+                    # Regular rounding for normal numbers
+                    repl[node] = Float(round(val, ndigits))
+
+        return e.xreplace(repl)
+
+    # Apply scientific rounding
+    expr_sci = scientific_round_expr(expr, 2)
+
+    # Convert to LaTeX
+    latex_eq_raw = latex(expr_sci, fold_frac_powers=True, fold_short_frac=True, mul_symbol='times')
+
+    # Replace safe names with readable display names
+    for name in feature_coefs.keys():
+        safe_name = name.replace(' ', '_').replace('^', '_pow_')
+        display_name = name.replace('_', ' ')
+        latex_eq_raw = latex_eq_raw.replace(safe_name, f"\\mathrm{{{display_name}}}")
+
+    # Add "y = " prefix
+    latex_eq = f"y = {latex_eq_raw}"
+
+    # Build Python format
+    python_parts = []
+    for name, coef in feature_coefs.items():
+        coef_str = format_number(coef)
+        if coef >= 0:
+            python_parts.append(f"+ {coef_str}*{name}")
+        else:
+            python_parts.append(f"- {format_number(abs(coef))}*{name}")
+
+    intercept_str_clean = format_number(intercept)
+    python_eq = f"y = {intercept_str_clean} " + ' '.join(python_parts) if python_parts else f"y = {intercept_str_clean}"
+
+    # Generate Excel format
+    col_letters = {name: chr(65 + i) for i, name in enumerate(feature_coefs.keys())}
+
+    excel_parts = []
+    for name, coef in feature_coefs.items():
+        coef_str = format_number(coef)
+        col_letter = col_letters[name]
+        if coef >= 0:
+            excel_parts.append(f"+ {coef_str}*{col_letter}1")
+        else:
+            excel_parts.append(f"- {format_number(abs(coef))}*{col_letter}1")
+
+    excel_eq = f"={intercept_str_clean} " + ' '.join(excel_parts) if excel_parts else f"={intercept_str_clean}"
+
+    return {
+        "latex": latex_eq,
+        "python": python_eq,
+        "excel": excel_eq
+    }
+
+def run_regression_analysis(df: pd.DataFrame, x_cols: List[str], y_col: str, model_type: str = "linear", poly_degree: int = 1, include_interactions: bool = False) -> Dict[str, Any]:
+    # 1. Prep Data
+    # Capture original X for plotting before transformation
+    X_original = df[x_cols].apply(pd.to_numeric, errors='coerce')
+    y_data = df[y_col]
+    
+    # Align indices after dropna
+    data = pd.concat([X_original, y_data], axis=1).dropna()
+    if data.empty or len(data) < len(x_cols) + 1:
+        raise ValueError("Insufficient data.")
+        
+    X_raw = data[x_cols] # Keep for plotting
+    y = pd.to_numeric(data[y_col], errors='coerce')
+    
+    X = X_raw.copy() # Start with raw for modelling
+
+    # 2. Advanced Feature Engineering
+    if model_type == "polynomial" or include_interactions:
+        degree = poly_degree if model_type == "polynomial" else 2
+        interaction_only = include_interactions and model_type != "polynomial"
+        poly = PolynomialFeatures(degree=degree, interaction_only=interaction_only, include_bias=False)
+        X_poly = poly.fit_transform(X)
+        poly_cols = poly.get_feature_names_out(X.columns)
+        X = pd.DataFrame(X_poly, columns=poly_cols, index=X.index)
+
+    # 3. Model Fitting
+    try:
+        model = None
+        y_pred = None
+        
+        if model_type == "logistic":
+            X_const = sm.add_constant(X)
+            y_bin = (y > y.median()).astype(int)
+            model = sm.Logit(y_bin, X_const).fit(disp=0)
+            y_pred = model.predict(X_const)
+            y = y_bin
+        elif model_type == "exponential":
+            if (y <= 0).any(): raise ValueError("Exponential regression requires Y > 0.")
+            y_log = np.log(y)
+            X_const = sm.add_constant(X)
+            lin_model = sm.OLS(y_log, X_const).fit()
+            y_pred = np.exp(lin_model.predict(X_const))
+            model = lin_model
+        else: # Linear or Polynomial
+            X_const = sm.add_constant(X)
+            model = sm.OLS(y, X_const).fit()
+            y_pred = model.predict(X_const)
+
+        # 4. Construct Visualization Data
+        # Create fit plots for each original feature
+        fit_plots_by_feature = {}
+        residuals_vs_fitted = []
+
+        y_list = y.tolist()
+        pred_list = y_pred.tolist()
+
+        residuals = []
+
+        # Create a fit plot for each original feature
+        for feature_name in X_raw.columns:
+            x_feature_list = X_raw[feature_name].tolist()
+            feature_plot = []
+
+            for i in range(len(y_list)):
+                feature_plot.append({
+                    "x": float(x_feature_list[i]),
+                    "real": float(y_list[i]),
+                    "pred": float(pred_list[i])
+                })
+
+            # Sort by X for proper curve rendering
+            feature_plot.sort(key=lambda item: item["x"])
+            fit_plots_by_feature[feature_name] = feature_plot
+
+        # Also create a single fit_plot using the first feature for backward compatibility
+        fit_plot = fit_plots_by_feature[X_raw.columns[0]] if len(X_raw.columns) > 0 else []
+
+        # Residuals plot
+        for i in range(len(y_list)):
+            res_val = y_list[i] - pred_list[i]
+            residuals.append(res_val)
+
+            residuals_vs_fitted.append({
+                "fitted": float(pred_list[i]),
+                "residual": res_val
+            })
+
+        # 5. Calculate Partial Regression Plots (Added Variable Plots)
+        # These show the isolated effect of each variable controlling for others
+        partial_regression_plots = {}
+
+        # Only calculate for multiple regression (more than 1 feature)
+        if len(X_raw.columns) > 1:
+            for feature_name in X_raw.columns:
+                # Get other features (all except current)
+                other_features = [col for col in X_raw.columns if col != feature_name]
+
+                if len(other_features) == 0:
+                    continue
+
+                # Step 1: Regress Y on all features except current one
+                X_other = X_raw[other_features]
+                X_other_const = sm.add_constant(X_other)
+                model_y = sm.OLS(y, X_other_const).fit()
+                y_residuals = y - model_y.predict(X_other_const)
+
+                # Step 2: Regress current feature on other features
+                model_x = sm.OLS(X_raw[feature_name], X_other_const).fit()
+                x_residuals = X_raw[feature_name] - model_x.predict(X_other_const)
+
+                # Step 3: Create partial plot data
+                partial_plot = []
+                for i in range(len(y)):
+                    partial_plot.append({
+                        "x": float(x_residuals.iloc[i]),
+                        "y": float(y_residuals.iloc[i])
+                    })
+
+                # Sort by x for proper line rendering
+                partial_plot.sort(key=lambda item: item["x"])
+                partial_regression_plots[feature_name] = partial_plot
+
+        # Generate equation strings
+        equations = generate_equations(model.params.to_dict(), model_type)
+
+        summary = {
+            "r_squared": float(model.rsquared) if hasattr(model, 'rsquared') else float(model.prsquared),
+            "adj_r_squared": float(model.rsquared_adj) if hasattr(model, 'rsquared_adj') else None,
+            "aic": float(model.aic),
+            "bic": float(model.bic),
+            "coefficients": model.params.to_dict(),
+            "p_values": model.pvalues.to_dict(),
+            "std_errors": model.bse.to_dict(),
+            "sample_size": int(model.nobs),
+            "residuals": residuals,
+            "fit_plot": fit_plot,  # Backward compatibility (first feature)
+            "fit_plots_by_feature": fit_plots_by_feature,  # All features
+            "partial_regression_plots": partial_regression_plots,  # Partial plots for multivariate
+            "diagnostic_plot": residuals_vs_fitted,
+            "equations": equations  # LaTeX, Python, Excel formats
+        }
+        return summary
+    except Exception as e:
+        raise ValueError(f"Model calculation failed: {str(e)}")

backend/generate_outlier_test_data.py

@@ -0,0 +1,294 @@
+#!/usr/bin/env python3
+"""
+Générateur de données de test pour détecter les outliers UNIVARIÉS et MULTIVARIÉS.
+
+Ce script crée un dataset avec :
+1. Des outliers univariés évidents (valeurs extrêmes dans une seule colonne)
+2. Des outliers multivariés (combinaisons de valeurs normales individuellement mais anormales ensemble)
+3. Des données normales pour la majorité
+"""
+
+import pandas as pd
+import numpy as np
+from pathlib import Path
+
+# Configuration
+np.random.seed(42)
+N_NORMAL = 100  # Nombre de lignes normales
+
+def generate_outlier_dataset():
+    """
+    Génère un dataset avec des outliers contrôlés pour tester les deux types de détection.
+    """
+    data = []
+
+    # ========================================================================
+    # 1. DONNÉES NORMALES (baseline)
+    # ========================================================================
+    print("📊 Génération des données normales...")
+
+    for i in range(N_NORMAL):
+        data.append({
+            "ID": i + 1,
+            "Age": np.random.normal(40, 10),  # Moyenne 40, écart-type 10
+            "Salaire": np.random.normal(35000, 8000),  # Moyenne 35k, écart-type 8k
+            "Experience": np.random.normal(10, 4),  # Moyenne 10 ans, écart-type 4
+            "Performance": np.random.normal(75, 10),  # Moyenne 75/100, écart-type 10
+            "Heures_Sup": np.random.normal(5, 3),  # Moyenne 5h/mois, écart-type 3
+            "Type": "Normal"
+        })
+
+    # ========================================================================
+    # 2. OUTLIERS UNIVARIÉS (évidents dans UNE colonne)
+    # ========================================================================
+    print("🔴 Génération des outliers univariés...")
+
+    # Outlier 1: Âge extrême (150 ans - impossible)
+    data.append({
+        "ID": 101,
+        "Age": 150,
+        "Salaire": 38000,
+        "Experience": 12,
+        "Performance": 78,
+        "Heures_Sup": 6,
+        "Type": "Outlier_Uni_Age"
+    })
+
+    # Outlier 2: Salaire extrêmement élevé (500k - 10x la normale)
+    data.append({
+        "ID": 102,
+        "Age": 45,
+        "Salaire": 500000,
+        "Experience": 15,
+        "Performance": 82,
+        "Heures_Sup": 8,
+        "Type": "Outlier_Uni_Salaire"
+    })
+
+    # Outlier 3: Salaire négatif (impossible)
+    data.append({
+        "ID": 103,
+        "Age": 35,
+        "Salaire": -5000,
+        "Experience": 8,
+        "Performance": 72,
+        "Heures_Sup": 4,
+        "Type": "Outlier_Uni_Salaire_Neg"
+    })
+
+    # Outlier 4: Performance > 100 (impossible)
+    data.append({
+        "ID": 104,
+        "Age": 38,
+        "Salaire": 42000,
+        "Experience": 11,
+        "Performance": 150,
+        "Heures_Sup": 7,
+        "Type": "Outlier_Uni_Perf"
+    })
+
+    # Outlier 5: Heures supplémentaires négatives
+    data.append({
+        "ID": 105,
+        "Age": 42,
+        "Salaire": 36000,
+        "Experience": 13,
+        "Performance": 76,
+        "Heures_Sup": -20,
+        "Type": "Outlier_Uni_Heures"
+    })
+
+    # ========================================================================
+    # 3. OUTLIERS MULTIVARIÉS (normaux individuellement, anormaux ensemble)
+    # ========================================================================
+    print("🟣 Génération des outliers multivariés...")
+
+    # Outlier Multivarié 1: Jeune avec BEAUCUP d'expérience (impossible)
+    # Age=25 (normal) mais Experience=30 (impossible pour cet âge)
+    data.append({
+        "ID": 201,
+        "Age": 25,
+        "Salaire": 32000,
+        "Experience": 30,
+        "Performance": 70,
+        "Heures_Sup": 5,
+        "Type": "Outlier_Multi_Age_Exp"
+    })
+
+    # Outlier Multivarié 2: Haut salaire avec basse performance (suspect)
+    # Salaire=80k (normal possible) mais Performance=40 (anormalement bas pour ce salaire)
+    data.append({
+        "ID": 202,
+        "Age": 45,
+        "Salaire": 80000,
+        "Experience": 15,
+        "Performance": 40,
+        "Heures_Sup": 2,
+        "Type": "Outlier_Multi_Salaire_Perf"
+    })
+
+    # Outlier Multivarié 3: Faible expérience avec très haut salaire (suspect)
+    data.append({
+        "ID": 203,
+        "Age": 28,
+        "Salaire": 95000,
+        "Experience": 1,
+        "Performance": 85,
+        "Heures_Sup": 15,
+        "Type": "Outlier_Multi_Exp_Salaire"
+    })
+
+    # Outlier Multivarié 4: Personne âgée avec junior-level tout
+    data.append({
+        "ID": 204,
+        "Age": 65,
+        "Salaire": 25000,
+        "Experience": 1,
+        "Performance": 60,
+        "Heures_Sup": 0,
+        "Type": "Outlier_Multi_Senior_Junior"
+    })
+
+    # Outlier Multivarié 5: Performance parfaite avec 0 heures supp (rare)
+    data.append({
+        "ID": 205,
+        "Age": 35,
+        "Salaire": 40000,
+        "Experience": 10,
+        "Performance": 100,
+        "Heures_Sup": 0,
+        "Type": "Outlier_Multi_Perf_Heures"
+    })
+
+    # Outlier Multivarié 6: Combinaison impossible - Junior avec salaire senior ET perf max
+    data.append({
+        "ID": 206,
+        "Age": 22,
+        "Salaire": 85000,
+        "Experience": 0,
+        "Performance": 95,
+        "Heures_Sup": 0,
+        "Type": "Outlier_Multi_Impossible"
+    })
+
+    # ========================================================================
+    # 4. CAS LIMITES (valeurs frontières)
+    # ========================================================================
+    print("🎯 Génération des cas limites...")
+
+    # Cas limite 1: Zéro partout (suspect mais pas impossible)
+    data.append({
+        "ID": 301,
+        "Age": 22,
+        "Salaire": 0,
+        "Experience": 0,
+        "Performance": 0,
+        "Heures_Sup": 0,
+        "Type": "Cas_Limie_Zeros"
+    })
+
+    # Cas limite 2: Très âgé avec beaucoup d'expérience (normal)
+    data.append({
+        "ID": 302,
+        "Age": 62,
+        "Salaire": 70000,
+        "Experience": 40,
+        "Performance": 88,
+        "Heures_Sup": 3,
+        "Type": "Normal_Senior"
+    })
+
+    # Cas limite 3: Salaire minimum légal (normal)
+    data.append({
+        "ID": 303,
+        "Age": 25,
+        "Salaire": 15000,
+        "Experience": 2,
+        "Performance": 65,
+        "Heures_Sup": 2,
+        "Type": "Normal_Salaire_Min"
+    })
+
+    # Création du DataFrame
+    df = pd.DataFrame(data)
+
+    # Arrondir les valeurs numériques pour plus de clarté
+    numeric_cols = ["Age", "Salaire", "Experience", "Performance", "Heures_Sup"]
+    for col in numeric_cols:
+        df[col] = df[col].round(2)
+
+    # Réordonner les colonnes
+    df = df[["ID", "Age", "Experience", "Salaire", "Performance", "Heures_Sup", "Type"]]
+
+    return df
+
+def main():
+    print("=" * 70)
+    print("🧪 GÉNÉRATEUR DE DONNÉES DE TEST - OUTLIERS UNIVARIÉS & MULTIVARIÉS")
+    print("=" * 70)
+    print()
+
+    # Générer le dataset
+    df = generate_outlier_dataset()
+
+    # Sauvegarder en CSV
+    output_dir = Path("/home/sepehr/dev/Data_analysis/backend/test_data")
+    output_dir.mkdir(exist_ok=True)
+
+    csv_path = output_dir / "test_outliers_complete.csv"
+    df.to_csv(csv_path, index=False)
+
+    excel_path = output_dir / "test_outliers_complete.xlsx"
+    df.to_excel(excel_path, index=False)
+
+    # Afficher les statistiques
+    print()
+    print("=" * 70)
+    print("📊 STATISTIQUES DU DATASET")
+    print("=" * 70)
+    print(f"✅ Total lignes : {len(df)}")
+    print(f"📈 Colonnes : {len(df.columns)}")
+    print()
+    print("🔴 Outliers univariés attendus : 5")
+    print("   - ID 101: Âge = 150 ans")
+    print("   - ID 102: Salaire = 500,000€")
+    print("   - ID 103: Salaire = -5,000€ (négatif)")
+    print("   - ID 104: Performance = 150 (>100)")
+    print("   - ID 105: Heures_Sup = -20 (négatif)")
+    print()
+    print("🟣 Outliers multivariés attendus : 6")
+    print("   - ID 201: Âge=25 avec Exp=30 (impossible)")
+    print("   - ID 202: Salaire=80k avec Perf=40 (incohérent)")
+    print("   - ID 203: Exp=1 avec Salaire=95k (suspect)")
+    print("   - ID 204: Âge=65 avec Exp=1 (incohérent)")
+    print("   - ID 205: Perf=100 avec Heures_Sup=0 (rare)")
+    print("   - ID 206: Âge=22, Exp=0, Salaire=85k (impossible)")
+    print()
+    print("=" * 70)
+    print("💾 FICHIERS GÉNÉRÉS")
+    print("=" * 70)
+    print(f"📄 CSV : {csv_path}")
+    print(f"📊 Excel : {excel_path}")
+    print()
+    print("=" * 70)
+    print("🎯 COMMENT TESTER")
+    print("=" * 70)
+    print("1. Importez le fichier 'test_outliers_complete.csv' dans l'application")
+    print("2. Vérifiez que les colonnes sont bien détectées comme numériques")
+    print("3. Les cercles ROUGES doivent apparaître sur les colonnes avec outliers univariés")
+    print("4. Le cercle VIOLET doit apparaître (indicateur global multivarié)")
+    print("5. Cliquez sur chaque indicateur pour voir les détails")
+    print("6. Vérifiez la cohérence des outliers détectés")
+    print()
+    print("✨ Bon testing !")
+    print("=" * 70)
+
+    # Afficher un aperçu des outliers
+    print()
+    print("📋 APERÇU DES OUTLIERS DANS LE DATASET :")
+    print("-" * 70)
+    outliers_df = df[df["Type"].str.contains("Outlier", case=False, na=False)]
+    print(outliers_df[["ID", "Type", "Age", "Experience", "Salaire", "Performance", "Heures_Sup"]].to_string(index=False))
+
+if __name__ == "__main__":
+    main()

backend/generate_test_data.py

@@ -0,0 +1,266 @@
+"""
+Générateur de données de test pour l'outil d'analyse de données.
+Crée des fichiers CSV et XLSX avec des corrélations et relations polynomiales.
+"""
+
+import pandas as pd
+import numpy as np
+from pathlib import Path
+
+# Configuration
+np.random.seed(42)
+n_samples = 500
+output_dir = Path(__file__).parent.parent / "test_data"
+output_dir.mkdir(exist_ok=True)
+
+print(f"📁 Génération des données de test dans: {output_dir}")
+
+# ============================================================================
+# 1. Dataset Ventes & Marketing (corrélations multiples, relations linéaires)
+# ============================================================================
+print("\n📊 Dataset: Ventes & Marketing")
+
+ventes_data = {
+    # Budget marketing mensuel (€)
+    'budget_marketing': np.random.uniform(1000, 50000, n_samples),
+
+    # Heures de publicité TV
+    'tv_ads_hours': np.random.uniform(5, 100, n_samples),
+
+    # Nombre de posts sur réseaux sociaux
+    'social_media_posts': np.random.randint(10, 200, n_samples),
+
+    # Prix moyen du produit (€) - légère corrélation négative avec ventes
+    'prix_moyen': np.random.uniform(20, 100, n_samples),
+
+    # Temperature moyenne (°C) - effet saisonnier
+    'temperature': np.random.normal(20, 8, n_samples),
+
+    # Concurrence (index 1-10)
+    'concurrence': np.random.randint(1, 11, n_samples),
+}
+
+df_ventes = pd.DataFrame(ventes_data)
+
+# Créer les ventes avec des relations réalistes
+# Ventes de base
+ventes_base = 5000
+
+# Effet budget marketing (fortement positif)
+effet_budget = df_ventes['budget_marketing'] * 0.8
+
+# Effet TV ads (modéré)
+effet_tv = df_ventes['tv_ads_hours'] * 25
+
+# Effet social media (faible mais positif)
+effet_social = df_ventes['social_media_posts'] * 8
+
+# Effet prix (négatif - prix plus élevé = moins de ventes)
+effet_prix = -df_ventes['prix_moyen'] * 15
+
+# Effet température (positif - meilleur temps = plus de ventes)
+effet_temp = df_ventes['temperature'] * 30
+
+# Effet concurrence (négatif)
+effet_concurrence = -df_ventes['concurrence'] * 200
+
+# Bruit aléatoire
+bruit = np.random.normal(0, 500, n_samples)
+
+# Ventes totales
+df_ventes['ventes'] = (ventes_base + effet_budget + effet_tv + effet_social +
+                      effet_prix + effet_temp + effet_concurrence + bruit)
+
+# Arrondir
+df_ventes['ventes'] = df_ventes['ventes'].round(2)
+
+# Sauvegarder
+ventes_csv = output_dir / "ventes_marketing.csv"
+ventes_excel = output_dir / "ventes_marketing.xlsx"
+
+df_ventes.to_csv(ventes_csv, index=False)
+df_ventes.to_excel(ventes_excel, index=False)
+
+print(f"   ✅ {ventes_csv.name}: {n_samples} lignes, {len(df_ventes.columns)} colonnes")
+print(f"   ✅ {ventes_excel.name}: {n_samples} lignes, {len(df_ventes.columns)} colonnes")
+
+# ============================================================================
+# 2. Dataset Production Industriel (relation polynomiale / quadratique)
+# ============================================================================
+print("\n🏭 Dataset: Production Industrielle")
+
+production_data = {
+    # Température machine (°C)
+    'temperature_machine': np.random.uniform(150, 250, n_samples),
+
+    # Pression (bar)
+    'pression': np.random.uniform(2, 10, n_samples),
+
+    # Vitesse conveyor (m/min)
+    'vitesse_conveyor': np.random.uniform(50, 150, n_samples),
+
+    # Humidité (%)
+    'humidite': np.random.uniform(30, 70, n_samples),
+
+    # Qualité matière première (index 1-100)
+    'qualite_matiere': np.random.uniform(60, 100, n_samples),
+}
+
+df_production = pd.DataFrame(production_data)
+
+# Relation polynomiale: température optimale ~200°C
+# Efficacité = a*(T - Toptimal)^2 + b
+T_optimal = 200
+df_production['efficacite_production'] = (
+    -0.08 * (df_production['temperature_machine'] - T_optimal) ** 2
+    + 95  # Efficacité maximale
+    + df_production['pression'] * 1.5
+    + df_production['vitesse_conveyor'] * 0.1
+    + df_production['qualite_matiere'] * 0.3
+    + np.random.normal(0, 3, n_samples)
+)
+
+# Borner entre 0 et 100
+df_production['efficacite_production'] = df_production['efficacite_production'].clip(0, 100).round(2)
+
+# Défauts de production (relation inverse avec l'efficacité)
+df_production['defauts'] = (
+    100 - df_production['efficacite_production']
+) * 0.5 + np.random.normal(0, 1, n_samples)
+df_production['defauts'] = df_production['defauts'].clip(0, None).round(2)
+
+# Sauvegarder
+production_csv = output_dir / "production_industrielle.csv"
+production_excel = output_dir / "production_industrielle.xlsx"
+
+df_production.to_csv(production_csv, index=False)
+df_production.to_excel(production_excel, index=False)
+
+print(f"   ✅ {production_csv.name}: {n_samples} lignes, {len(df_production.columns)} colonnes")
+print(f"   ✅ {production_excel.name}: {n_samples} lignes, {len(df_production.columns)} colonnes")
+
+# ============================================================================
+# 3. Dataset Santé & Fitness (relations mixtes)
+# ============================================================================
+print("\n🏃 Dataset: Santé & Fitness")
+
+sante_data = {
+    # Âge (années)
+    'age': np.random.randint(18, 80, n_samples),
+
+    # Poids (kg)
+    'poids': np.random.normal(75, 15, n_samples),
+
+    # Taille (cm)
+    'taille': np.random.normal(170, 10, n_samples),
+
+    # Heures d'exercice par semaine
+    'heures_exercice': np.random.uniform(0, 15, n_samples),
+
+    # Calories consommées par jour
+    'calories_jour': np.random.normal(2200, 400, n_samples),
+
+    # Heures de sommeil
+    'heures_sommeil': np.random.normal(7, 1.5, n_samples),
+
+    # Niveau de stress (1-10)
+    'stress': np.random.randint(1, 11, n_samples),
+}
+
+df_sante = pd.DataFrame(sante_data)
+
+# IMC (BMI)
+df_sante['imc'] = (df_sante['poids'] / (df_sante['taille'] / 100) ** 2).round(2)
+
+# Métabolisme de base (formule de Harris-Benedict simplifiée)
+# Hommes: BMR = 88.362 + (13.397 × kg) + (4.799 × cm) - (5.677 × age)
+df_sante['metabolisme_base'] = (
+    88.362
+    + 13.397 * df_sante['poids']
+    + 4.799 * df_sante['taille']
+    - 5.677 * df_sante['age']
+    + np.random.normal(0, 50, n_samples)
+).round(2)
+
+# Niveau d'énergie (subjectif 1-10)
+df_sante['niveau_energie'] = (
+    5
+    + 0.3 * df_sante['heures_exercice']
+    - 0.2 * df_sante['stress']
+    + 0.15 * df_sante['heures_sommeil']
+    - 0.01 * (df_sante['age'] - 30)
+    + np.random.normal(0, 1, n_samples)
+).clip(1, 10).round(2)
+
+# Sauvegarder
+sante_csv = output_dir / "sante_fitness.csv"
+sante_excel = output_dir / "sante_fitness.xlsx"
+
+df_sante.to_csv(sante_csv, index=False)
+df_sante.to_excel(sante_excel, index=False)
+
+print(f"   ✅ {sante_csv.name}: {n_samples} lignes, {len(df_sante.columns)} colonnes")
+print(f"   ✅ {sante_excel.name}: {n_samples} lignes, {len(df_sante.columns)} colonnes")
+
+# ============================================================================
+# 4. Dataset Finance (relation exponentielle)
+# ============================================================================
+print("\n💰 Dataset: Finance & Investissement")
+
+finance_data = {
+    # Montant investi (€)
+    'montant_investi': np.random.uniform(1000, 100000, n_samples),
+
+    # Durée investissement (années)
+    'duree_annees': np.random.uniform(1, 30, n_samples),
+
+    # Taux de rendement annuel (%)
+    'taux_rendement': np.random.uniform(2, 15, n_samples),
+
+    # Risque (1-10)
+    'niveau_risque': np.random.randint(1, 11, n_samples),
+}
+
+df_finance = pd.DataFrame(finance_data)
+
+# Valeur finale avec intérêts composés: A = P(1 + r)^t
+df_finance['valeur_finale'] = (
+    df_finance['montant_investi'] *
+    (1 + df_finance['taux_rendement'] / 100) ** df_finance['duree_annees']
+    * (1 - 0.02 * df_finance['niveau_risque'])  # Pénalité de risque
+    + np.random.normal(0, df_finance['montant_investi'] * 0.01, n_samples)
+).round(2)
+
+# Profit
+df_finance['profit'] = (df_finance['valeur_finale'] - df_finance['montant_investi']).round(2)
+
+# Rendement total (%)
+df_finance['rendement_total'] = (df_finance['profit'] / df_finance['montant_investi'] * 100).round(2)
+
+# Sauvegarder
+finance_csv = output_dir / "finance_investissement.csv"
+finance_excel = output_dir / "finance_investissement.xlsx"
+
+df_finance.to_csv(finance_csv, index=False)
+df_finance.to_excel(finance_excel, index=False)
+
+print(f"   ✅ {finance_csv.name}: {n_samples} lignes, {len(df_finance.columns)} colonnes")
+print(f"   ✅ {finance_excel.name}: {n_samples} lignes, {len(df_finance.columns)} colonnes")
+
+# ============================================================================
+# Résumé
+# ============================================================================
+print("\n" + "="*60)
+print("✅ Tous les fichiers de test ont été générés avec succès !")
+print("="*60)
+print(f"\n📂 Répertoire: {output_dir}")
+print("\n📋 Fichiers créés:")
+print("   1. ventes_marketing.csv/xlsx    - Corrélations multiples linéaires")
+print("   2. production_industrielle.csv/xlsx - Relation polynomiale (quadratique)")
+print("   3. sante_fitness.csv/xlsx       - Relations mixtes + IMC calculé")
+print("   4. finance_investissement.csv/xlsx - Relation exponentielle")
+print("\n💡 Utilisez ces fichiers pour tester:")
+print("   • Analyse de corrélation")
+print("   • Régression linéaire, polynomiale, exponentielle")
+print("   • Import CSV et Excel")
+print("="*60)

backend/main.py

@@ -0,0 +1,32 @@
+from fastapi import FastAPI
+from fastapi.middleware.cors import CORSMiddleware
+from app.api.v1.upload import router as upload_router
+from app.api.v1.analysis import router as analysis_router
+from app.api.v1.reports import router as reports_router
+
+app = FastAPI(title="Data_analysis API", version="0.1.0")
+
+# CORS configuration
+origins = ["*"]  # Allow all origins for dev/homelab simplicity
+
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=origins,
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+    expose_headers=["X-Column-Metadata", "Content-Disposition"]
+)
+
+# Register routers
+app.include_router(upload_router, prefix="/api/v1")
+app.include_router(analysis_router, prefix="/api/v1")
+app.include_router(reports_router, prefix="/api/v1")
+
+@app.get("/health")
+def health_check():
+    return {"status": "ok", "service": "backend"}
+
+@app.get("/")
+def read_root():
+    return {"message": "Welcome to Data_analysis API"}

backend/pyproject.toml

@@ -0,0 +1,22 @@
+[project]
+name = "backend"
+version = "0.1.0"
+description = "Add your description here"
+readme = "README.md"
+requires-python = ">=3.12"
+dependencies = [
+    "fastapi>=0.128.0",
+    "fpdf2>=2.8.5",
+    "httpx>=0.28.1",
+    "matplotlib>=3.10.0",
+    "openpyxl>=3.1.5",
+    "pandas>=2.3.3",
+    "pyarrow>=22.0.0",
+    "pydantic>=2.12.5",
+    "pytest>=9.0.2",
+    "python-multipart>=0.0.21",
+    "scikit-learn>=1.8.0",
+    "statsmodels>=0.14.6",
+    "sympy>=1.13.0",
+    "uvicorn>=0.40.0",
+]

backend/test_data/test_outliers_complete.csv

@@ -0,0 +1,115 @@
+ID,Age,Experience,Salaire,Performance,Heures_Sup,Type
+1,44.97,12.59,33893.89,90.23,4.3,Normal
+2,37.66,13.07,47633.7,70.31,6.63,Normal
+3,35.37,10.97,31274.16,55.87,-0.17,Normal
+4,34.38,11.26,26897.35,65.92,0.76,Normal
+5,54.66,10.27,33193.79,60.75,3.37,Normal
+6,41.11,11.5,25792.05,68.99,4.12,Normal
+7,33.98,9.95,49818.23,64.42,7.47,Normal
+8,27.79,2.16,36670.91,61.72,5.59,Normal
+9,47.38,9.54,36370.95,71.99,0.56,Normal
+10,32.8,14.23,31314.89,78.44,-0.29,Normal
+11,43.24,7.29,31919.34,81.12,8.09,Normal
+12,49.31,8.76,28286.26,78.31,7.93,Normal
+13,35.21,5.57,33514.73,63.04,7.44,Normal
+14,53.56,14.01,34423.92,78.62,3.06,Normal
+15,43.61,9.86,47304.29,90.65,-2.86,Normal
+16,48.22,8.8,35696.38,75.92,-0.96,Normal
+17,37.8,15.91,37856.9,69.82,2.57,Normal
+18,34.98,11.32,42323.22,69.7,6.54,Normal
+19,40.97,7.19,42749.16,71.72,3.82,Normal
+20,25.36,11.04,37368.96,75.05,4.3,Normal
+21,25.85,8.63,31634.84,66.98,4.52,Normal
+22,44.04,10.7,50089.49,77.58,4.78,Normal
+23,20.81,10.24,34787.89,99.63,4.42,Normal
+24,43.02,5.33,34722.31,86.43,7.26,Normal
+25,47.91,15.61,27724.9,60.98,6.76,Normal
+26,61.9,7.73,27075.71,76.0,3.49,Normal
+27,24.49,5.75,35548.5,79.74,2.24,Normal
+28,55.5,8.71,28733.97,83.14,1.31,Normal
+29,42.27,3.57,45457.14,76.85,5.78,Normal
+30,47.82,4.72,25104.39,80.22,5.89,Normal
+31,42.5,7.28,37771.59,77.32,5.88,Normal
+32,32.86,11.9,49926.2,63.09,6.97,Normal
+33,30.25,14.63,41296.68,66.79,7.89,Normal
+34,44.13,17.59,41576.48,72.55,2.74,Normal
+35,31.1,9.69,28473.52,78.41,5.83,Normal
+36,48.27,15.81,35104.02,72.35,13.16,Normal
+37,46.26,5.72,28142.74,79.82,4.33,Normal
+38,47.14,9.71,38785.9,66.53,0.46,Normal
+39,35.53,10.86,41851.19,62.54,5.52,Normal
+40,43.85,10.61,27929.14,75.58,1.57,Normal
+41,43.58,14.33,39486.28,85.54,0.87,Normal
+42,30.62,12.06,39120.28,80.15,16.56,Normal
+43,45.71,13.82,44084.53,81.51,4.05,Normal
+44,47.59,9.05,28817.4,70.15,5.25,Normal
+45,63.15,12.75,20061.88,58.87,3.58,Normal
+46,50.89,5.69,35514.24,67.85,7.04,Normal
+47,32.7,10.18,36731.67,68.48,11.43,Normal
+48,46.34,10.75,18798.86,68.38,7.56,Normal
+49,32.07,12.02,34082.11,83.66,1.4,Normal
+50,36.65,7.39,31200.44,92.65,6.21,Normal
+51,27.39,18.49,42342.9,85.32,0.44,Normal
+52,35.16,7.17,45135.29,79.44,7.32,Normal
+53,30.73,-2.97,34523.8,64.76,4.24,Normal
+54,27.52,4.28,48059.29,70.6,5.39,Normal
+55,54.41,14.65,23513.1,75.1,2.06,Normal
+56,44.62,7.6,36592.48,75.7,3.84,Normal
+57,41.14,16.34,40297.05,62.62,11.4,Normal
+58,20.48,12.35,33785.72,77.81,3.13,Normal
+59,37.92,7.64,31055.99,83.5,6.07,Normal
+60,33.07,11.23,42196.8,83.13,6.89,Normal
+61,31.71,12.99,30518.55,81.1,4.94,Normal
+62,41.17,7.63,45221.32,80.47,4.39,Normal
+63,37.82,13.3,43790.21,83.14,8.92,Normal
+64,40.21,8.76,40455.62,78.24,4.61,Normal
+65,40.97,6.73,39761.26,95.92,1.98,Normal
+66,27.86,13.17,44264.89,81.24,6.89,Normal
+67,39.88,10.3,27821.97,68.23,7.93,Normal
+68,38.53,8.71,28396.02,79.13,3.31,Normal
+69,31.78,10.98,36949.5,69.93,3.59,Normal
+70,42.32,4.37,23415.33,67.82,4.36,Normal
+71,43.11,13.43,46802.85,73.4,4.94,Normal
+72,29.97,8.85,34851.89,78.23,2.52,Normal
+73,45.19,9.56,47261.91,79.02,7.07,Normal
+74,35.99,10.05,36792.74,75.98,2.68,Normal
+75,40.25,15.8,38983.99,84.59,11.46,Normal
+76,32.33,10.73,41978.57,96.9,2.58,Normal
+77,31.6,1.5,30204.86,69.74,2.72,Normal
+78,41.5,17.5,37734.05,84.5,3.27,Normal
+79,31.02,4.72,38935.35,93.31,8.54,Normal
+80,35.31,15.42,21294.92,73.85,8.71,Normal
+81,24.06,10.02,30205.0,75.47,3.65,Normal
+82,46.23,9.43,26459.04,76.2,6.54,Normal
+83,47.12,3.86,26002.86,87.78,6.0,Normal
+84,32.52,10.46,47409.22,86.79,5.2,Normal
+85,60.61,9.0,49042.73,84.72,6.94,Normal
+86,53.69,12.74,27280.61,85.58,-0.28,Normal
+87,28.17,8.92,18686.14,82.18,9.51,Normal
+88,40.74,4.48,48028.92,57.97,4.83,Normal
+89,43.84,1.73,34738.44,74.11,1.09,Normal
+90,46.7,6.24,37932.79,69.86,1.82,Normal
+91,39.37,6.06,42641.14,80.04,3.41,Normal
+92,32.07,5.86,34143.76,69.46,1.41,Normal
+93,59.65,7.2,35282.11,77.14,4.66,Normal
+94,37.79,13.03,39913.33,69.69,3.27,Normal
+95,37.25,3.94,16584.63,88.67,9.93,Normal
+96,37.51,11.25,39612.46,105.79,8.36,Normal
+97,38.72,3.57,27355.68,77.03,2.73,Normal
+98,25.78,5.67,29827.42,91.87,7.64,Normal
+99,39.92,10.31,46839.55,66.39,9.57,Normal
+100,45.39,9.24,26702.03,66.24,0.85,Normal
+101,150.0,12.0,38000.0,78.0,6.0,Outlier_Uni_Age
+102,45.0,15.0,500000.0,82.0,8.0,Outlier_Uni_Salaire
+103,35.0,8.0,-5000.0,72.0,4.0,Outlier_Uni_Salaire_Neg
+104,38.0,11.0,42000.0,150.0,7.0,Outlier_Uni_Perf
+105,42.0,13.0,36000.0,76.0,-20.0,Outlier_Uni_Heures
+201,25.0,30.0,32000.0,70.0,5.0,Outlier_Multi_Age_Exp
+202,45.0,15.0,80000.0,40.0,2.0,Outlier_Multi_Salaire_Perf
+203,28.0,1.0,95000.0,85.0,15.0,Outlier_Multi_Exp_Salaire
+204,65.0,1.0,25000.0,60.0,0.0,Outlier_Multi_Senior_Junior
+205,35.0,10.0,40000.0,100.0,0.0,Outlier_Multi_Perf_Heures
+206,22.0,0.0,85000.0,95.0,0.0,Outlier_Multi_Impossible
+301,22.0,0.0,0.0,0.0,0.0,Cas_Limie_Zeros
+302,62.0,40.0,70000.0,88.0,3.0,Normal_Senior
+303,25.0,2.0,15000.0,65.0,2.0,Normal_Salaire_Min

backend/tests/test_analysis.py

@@ -0,0 +1,80 @@
+from fastapi.testclient import TestClient
+from main import app
+
+client = TestClient(app)
+
+def test_outlier_detection_univariate():
+    # Data with one clear outlier (100)
+    data = [
+        {"val": 10}, {"val": 11}, {"val": 12}, {"val": 10}, {"val": 100}
+    ]
+    request_data = {
+        "data": data,
+        "columns": ["val"],
+        "method": "univariate"
+    }
+    
+    response = client.post("/api/v1/analysis/detect-outliers", json=request_data)
+    assert response.status_code == 200
+    json_resp = response.json()
+    assert json_resp["total_count"] == 1
+    assert json_resp["outliers"][0]["index"] == 4
+    assert "IQR bounds" in json_resp["outliers"][0]["reasons"][0]
+
+def test_outlier_detection_multivariate():
+    # Data with a multivariate anomaly
+    data = [
+        {"x": 1, "y": 1}, {"x": 1.1, "y": 0.9}, {"x": 0.9, "y": 1.1},
+        {"x": 10, "y": 10} # Anomaly
+    ]
+    request_data = {
+        "data": data,
+        "columns": ["x", "y"],
+        "method": "multivariate"
+    }
+    
+    response = client.post("/api/v1/analysis/detect-outliers", json=request_data)
+    assert response.status_code == 200
+    json_resp = response.json()
+    assert json_resp["total_count"] >= 1
+
+def test_feature_importance():
+    data = [
+        {"x1": 1, "x2": 10, "y": 2},
+        {"x1": 2, "x2": 20, "y": 4},
+        {"x1": 3, "x2": 30, "y": 6},
+        {"x1": 4, "x2": 40, "y": 8},
+        {"x1": 5, "x2": 50, "y": 10}
+    ]
+    request_data = {
+        "data": data,
+        "features": ["x1", "x2"],
+        "target": "y"
+    }
+    response = client.post("/api/v1/analysis/feature-importance", json=request_data)
+    assert response.status_code == 200
+    json_resp = response.json()
+    assert len(json_resp["importances"]) == 2
+    assert "feature" in json_resp["importances"][0]
+    assert "score" in json_resp["importances"][0]
+
+def test_run_regression():
+    data = [
+        {"x": 1, "y": 2.1},
+        {"x": 2, "y": 3.9},
+        {"x": 3, "y": 6.2},
+        {"x": 4, "y": 8.1},
+        {"x": 5, "y": 10.3}
+    ]
+    request_data = {
+        "data": data,
+        "x_features": ["x"],
+        "y_target": "y",
+        "model_type": "linear"
+    }
+    response = client.post("/api/v1/analysis/run-regression", json=request_data)
+    assert response.status_code == 200
+    json_resp = response.json()
+    assert "results" in json_resp
+    assert json_resp["results"]["r_squared"] > 0.9
+    assert "const" in json_resp["results"]["coefficients"]

backend/tests/test_upload.py

@@ -0,0 +1,37 @@
+from fastapi.testclient import TestClient
+import pandas as pd
+import io
+import pyarrow as pa
+from main import app
+
+client = TestClient(app)
+
+def test_health_check():
+    response = client.get("/health")
+    assert response.status_code == 200
+    assert response.json() == {"status": "ok", "service": "backend"}
+
+def test_upload_csv():
+    # Create a dummy CSV
+    csv_content = "name,age\nAlice,30\nBob,25"
+    file = ("test.csv", csv_content, "text/csv")
+    
+    response = client.post("/api/v1/upload", files={"file": file})
+    
+    assert response.status_code == 200
+    assert response.headers["content-type"] == "application/vnd.apache.arrow.stream"
+    assert "X-Column-Metadata" in response.headers
+    
+    # Verify Arrow data
+    buffer = io.BytesIO(response.content)
+    with pa.ipc.open_stream(buffer) as reader:
+        table = reader.read_all()
+        df = table.to_pandas()
+        assert df.shape == (2, 2)
+        assert list(df.columns) == ["name", "age"]
+
+def test_upload_invalid_format():
+    file = ("test.txt", "invalid content", "text/plain")
+    response = client.post("/api/v1/upload", files={"file": file})
+    assert response.status_code == 400
+    assert "Only .xlsx, .xls and .csv files are supported" in response.json()["detail"]