Summary Dashboard
00 -- ML Portfolio Summary Dashboard (AUC comparison, score distributions, province analysis)
RTS Risk Classification (5 classifiers + ensemble)
01 -- ROC Curves: Logistic Regression | Random Forest | XGBoost | LightGBM | Decision Tree
02 -- XGBoost Feature Importances (top 15 features for RTS prediction)
03 -- Best Classifier Confusion Matrix (actual vs predicted RTS/Collected)
05 -- SVM (LinearSVC) Confusion Matrix
Neural Network
04 -- MLP Training Loss Curve + Validation Score
Delivery Time Regression (Ridge Regression)
06 -- Ridge Regression: Residuals vs Predicted + Residual Distribution (Transit Hours)
Customer Segmentation (K-Means)
07 -- Elbow Plot + Silhouette Scores (k=2 to k=8)
08 -- Customer Segments in PCA 2D Space (k=5 clusters)
09 -- Segment Profile Heatmap (normalised feature means per segment)
Anomaly Detection (Isolation Forest)
10 -- Anomaly Score Distribution with 5th Percentile Threshold
11 -- Anomaly vs Normal Feature Distributions (Value | Transit | Exceptions)
Return Reason Classification (Naive Bayes)
12 -- Naive Bayes Confusion Matrix (Low Risk | Dwell Risk | Exception Risk)
Demand Forecasting (ARIMA / Holt-Winters + Prophet Decomposition)
13 -- Holt-Winters Forecast (30 months train, 6 months forecast with 95% CI)
14 -- Prophet-Style Decomposition: Observed | Trend | Seasonal (12-month cycle)
Customer Churn Prediction (LightGBM)
15 -- LightGBM Churn Feature Importances
16 -- Calibration Curve: Predicted Probability vs Actual Fraction of Positives
Customer Lifetime Value (CLV) -- XGBoost Regressor
17 -- CLV Distributions: Historical vs Annualised vs Predicted (1yr)
18 -- Mean CLV by Province (bar + map overlay)
19 -- CLV Decile Analysis: top 20% customers = 69.9% of total value
20 -- CLV vs Churn Risk Matrix (tier segmentation)
21 -- XGBoost CLV Feature Importance (R²=0.987, MAE=R50)
Geographical Analysis -- SA Province Map
22 -- SA Province Customer Density Map + CLV vs RTS Rate by Province
23 -- Province Performance Dashboard: Customer Count | Mean CLV | RTS Rate
Correlation & Statistical Analysis
30 -- Full Feature Correlation Matrix (Pearson r, all parcel & customer attributes)
31 -- Feature vs RTS Target Correlation Bar Chart (which features drive RTS)
32 -- Scatter Matrix: Top 4 Predictive Features vs RTS (Collected vs RTS colour-coded)
33 -- Province x Feature Heatmap: Mean values by province with RTS % column
34 -- RTS Rate by Retailer Category and Province (with significance bands)
35 -- Feature Distributions: RTS vs Collected (KS test statistic per feature)
36 -- CLV Correlations: Feature impact on Historical vs Predicted 1-Year CLV
37 -- Time-Series Correlation: ACF, Seasonal Volume, Lag-12 Scatter
38 -- Multi-Feature Relationship Scatter Matrix (colour = RTS risk, trend line = Pearson r)
All 15 Models -- Inventory