How luu-quang/data_mining_and_machine_learning_usth works
This is a batch banking-product recommendation workflow: customer CSV data is converted, cleaned, transformed into temporal product-addition targets, and used to rank new-product recommendations with out-of-time MAP@5 evaluation. The README also describes EDA, feature engineering, optional modeling approaches, and optimization; the available scripts concretely implement conversion, cleaning, PCA exploration, and a random-forest pipeline. Edges reflect only script calls and documented workflow order; the scripts are standalone and use hard-coded file paths, so links between stages are documented rather than direct module calls.
Main entry points: 01_eda_and_cleaning.py Cleaning and imputation · 00_convert_format.py Format conversion · 03_finalize_model.py Temporal features and targets · 02_feature_extraction_pca.py PCA exploration
Components 12
Data preparation
- Format conversion
00_convert_format.py - Cleaning and imputation
01_eda_and_cleaning.py - Optimized Parquet data
00_convert_format.py - Cleaned Parquet data
01_eda_and_cleaning.py
Feature engineering
- Behavior exploration
01_eda_and_cleaning.py - PCA exploration
02_feature_extraction_pca.py - Temporal features and targets
03_finalize_model.py - Scaled PCA transform
03_finalize_model.py
Recommendation modeling
- Product-specific forests
03_finalize_model.py - Ranked recommendations
03_finalize_model.py
Evaluation
- MAP@5 evaluation
03_finalize_model.py
Other components
- Bank customer data
Connections 11
- Bank customer data → Format conversion: loads CSV
00_convert_format.py - Format conversion → Optimized Parquet data: writes Parquet
00_convert_format.py - Optimized Parquet data → Cleaning and imputation: reads Parquet
README.md - Cleaning and imputation → Cleaned Parquet data: writes Parquet
01_eda_and_cleaning.py - Cleaned Parquet data → Temporal features and targets: loads datasets
03_finalize_model.py - Cleaning and imputation → Behavior exploration: supports exploration
README.md - Cleaned Parquet data → PCA exploration: loads training data
02_feature_extraction_pca.py - Temporal features and targets → Scaled PCA transform: prepares split features
03_finalize_model.py - Scaled PCA transform → Product-specific forests: provides components
03_finalize_model.py - Product-specific forests → Ranked recommendations: scores and ranks
03_finalize_model.py - Ranked recommendations → MAP@5 evaluation: compared with purchases
03_finalize_model.py