[DL-04] Add CatBoost discount-depth regression model

[azharmujeeb]

Summary

Implements CatBoost regression for the DL-04 Price & Discount Forecasting feature.

Completed

• Converted discount prediction from yes/no classification to discount-depth regression.
• Prepared and cleaned the grocery dataset.
• Created leakage-safe train, validation and test splits without SKU overlap.
• Trained and tuned a CatBoostRegressor model.
• Completed feature analysis, error analysis and hyperparameter tuning.
• Final test results:
  • MAE: 2.36
  • RMSE: 4.88
  • R²: 0.84
• Saved model metadata and evaluation plots.
• Added a standardised JSON prediction prototype for future API integration.

Current limitation

The current training dataset contains only one timestamp, so this model predicts discount depth from product features and is not yet a true future time-series price forecasting model.

Next steps

• Adapt the pipeline to a proper historical price-over-time dataset.
• Align the JSON/API structure with the Backend/App Development team.
• Integrate the CatBoost model into the DiscountMate ML service.