Bank Client Subscription Classifier

This project's primary goal is to demonstrate MLflow workflows for experiment tracking, model registry, reproducible ML pipelines, and hands-on familiarity with the MLflow UI and processes, while participating in Kaggle's Playground Series S5E8 competition.

Documentation

Full project analysis and documentation

Navigation Guide

The documentation follows the complete MLFlow journey through different versions (V1, V2, V3, V4) showing the evolution of the process. Navigate sequentially through the left navigation panel from top to bottom to follow the development progression and understand how each iteration builds upon the previous version.

Competition Details:

• Competition overview: Binary Classification with a Bank Dataset
• Dataset: S5E8 synthetic banking data
• Task: Binary classification predicting term deposit subscriptions
• Evaluation: ROC AUC score
• Timeline: August 2025

Quick Setup

# Install dependencies
pip install -e .

# Start MLflow tracking server
mlflow server --backend-store-uri sqlite:///mlflow.db --default-artifact-root ./mlartifacts
# MLflow UI will be available at http://localhost:5000

# Run MKDocs
mkdocs serve

Project Structure

├── .github/                  # GitHub configuration
├── catboost_info/            # CatBoost training artifacts
├── config/                   # Configuration files and loaders
├── data/                     # Data files (raw, processed, predictions)
│   ├── predictions/          # Model predictions output
│   ├── processed/            # Processed datasets
│   └── raw/                  # Raw competition data
├── docs/                     # MkDocs documentation
├── images/                   # Project images and assets
├── mlartifacts/              # MLflow artifacts storage
├── mlruns/                   # MLflow experiment tracking data
├── notebooks/                # Jupyter notebooks for EDA and experiments
├── src/                      # Source code
│   ├── data/                 # Data loading and preprocessing modules
│   └── models/               # Model training, evaluation, and prediction
│       ├── core/             # Core model abstractions
│       └── implementations/  # Model implementations (LightGBM, XGBoost, CatBoost)
├── tests/                    # Unit tests
├── main.py                   # Main entry point
├── MLproject                 # MLflow project definition
├── mkdocs.yml                # MkDocs configuration
├── pyproject.toml            # Python project configuration
└── python_env.yaml           # Conda environment specification (Needed for MLFlow)

Development Status

The core project structure is complete with full MLflow implementation across multiple iterations (V1-V4). The Kaggle competition has concluded, but the project remains open for further enhancements such as SHAP values for feature importance analysis, additional model interpretability tools, and extended MLflow workflow demonstrations.