Bank Term Deposit Subscription Prediction

This project delivers an end-to-end machine learning solution for predicting whether a bank client will subscribe to a term deposit. By leveraging advanced modeling techniques, this predictor enables targeted marketing campaigns, optimizes resource allocation, and provides data-driven insights for strategic decision-making.

Live: https://banktermdepositsubscriptionprediction.streamlit.app/

---

Key Features & Technical Highlights

• End-to-End ML Pipeline: A complete, reproducible workflow from raw data ingestion to a saved, production-ready model using scikit-learn and imblearn pipelines.
• Advanced Feature Engineering: Creation of interaction features (e.g., balance_per_age) and meaningful ordinal encoding for features like education.
• Robust Imbalance Handling: Utilization of SMOTETomek to intelligently oversample the minority class and clean the decision boundary, leading to a more balanced and high-performing model.
• Comprehensive Model Evaluation: Benchmarking of baseline models against advanced gradient boosting models (LightGBM, XGBoost, CatBoost).
• Sophisticated Ensembling: Implementation of a Stacking Classifier that uses the tuned gradient boosting models as base learners to achieve superior and more robust performance.
• GPU Acceleration: The entire tuning and training process is optimized to run on a GPU, drastically reducing computation time from hours to minutes.
• Model Explainability: In-depth model interpretation using SHAP (SHapley Additive exPlanations) to understand the key drivers behind the final model's predictions.
• Ready for Deployment: The project concludes with a saved, unified pipeline and a complete Streamlit application script (app.py) for live predictions.

---

Tech Stack

Category	Technologies & Libraries
Data Manipulation	Python, Pandas, NumPy
Data Visualization	Matplotlib, Seaborn
ML Pipeline & Modeling	Scikit-learn, Imbalanced-learn, LightGBM, XGBoost, CatBoost
Explainability	SHAP
Deployment	Streamlit, Joblib
Development	Jupyter Notebook, Kaggle Notebooks, Google Colab

---

Project Workflow

The project follows a structured workflow designed to ensure robustness and prevent data leakage:

Data Loading → Feature Engineering → Data Splitting (80/20) → Preprocessing Pipeline (Fit on Train) → Resampling (on Train) → Hyperparameter Tuning (CV on Resampled Train) → Stacking Ensemble → Final Evaluation → SHAP Analysis → Save Artifacts

---

Final Model Performance

The models were evaluated on the 20% held-out test set after optimizing the prediction threshold to maximize the F1-Score. The leaderboard below shows that the tuned LightGBM model achieved the highest F1-Score in this run, closely followed by the other advanced models.

Model	Accuracy	Precision (Yes)	Recall (Yes)	F1-Score (Yes)	ROC-AUC
LightGBM	0.9325	0.6949	0.7856	0.7375	0.9674
XGBoost	0.9292	0.6786	0.7844	0.7277	0.9650
CatBoost	0.9280	0.6701	0.7940	0.7268	0.9648
Stacking Ensemble	0.9286	0.6757	0.7853	0.7264	0.9628

---

Model Explainability with SHAP

The SHAP summary plot reveals the most influential features driving the model's predictions for the positive class ("Yes" - will subscribe).

Key Insights:

• duration: The duration of the last contact is by far the most significant predictor. Longer calls are strongly associated with a higher likelihood of subscription.
• poutcome_success: A successful outcome in a previous campaign is a very strong positive indicator.
• Financials & Demographics: Features like balance, age, and housing also play important roles, with their impact varying for each individual prediction.

---

Repository Structure

bank-term-deposit-prediction/
│
├── 📂 data/
│   └── train.csv
│
├── 📂 saved_models/
│   ├── stacking_model_final_gpu.joblib
│   └── preprocessor_final_gpu.joblib
│
├── 📜 Bank.ipynb
├── 📜 Model.ipynb  
├── 📜 app.py                  
├── 📜 requirements.txt      
└── 📜 README.md

---

Setup and Installation

Follow these steps to set up the project environment.

1. Clone the Repository

git clone [https://github.com/itz-Mayank/Bank_Term_Deposit_Subscription_Prediction](https://github.com/itz-Mayank/Bank_Term_Deposit_Subscription_Prediction)]
cd Bank_Term_Deposit_Subscription_Prediction

2. Create a Python Environment

It is highly recommended to use a virtual environment (e.g., Conda) to manage dependencies.

conda create -n bank_predictor python=3.10 -y
conda activate bank_predictor

3. Install Dependencies

Install all the required libraries from the requirements.txt file.

pip install -r requirements.txt

Usage

There are two main ways to use this repository:

1. Training the Model

To re-run the entire training, tuning, and evaluation process, open and run the main notebook:

jupyter lab main_notebook.ipynb

2. Running the Prediction App To launch the user-facing web application for live predictions, ensure you have the saved model files (e.g., final_model_pipeline.joblib) in the root directory and run:

streamlit run app.py

Author

Created by Mayank Meghwal