Predicting Bank Customer Churn Using Machine Learning

Module: 25LLP132 – Principles of Artificial Intelligence and Data Analytics
Group: Creative Finance
Project: Predicting Bank Customer Churn Using Machine Learning
Dataset: Kaggle “Churn Modelling” (Churn_Modelling.csv)

This repository contains the reproducible pipeline used in our project .

Business Context

Customer churn is costly: retaining customers is usually cheaper than acquiring new ones.
Our goal is to identify high-risk customers early so the bank can apply targeted retention strategies.

Dataset Summary

• Rows: 10,000 customers
• Target: Exited (1 = churned, 0 = retained)
• Removed non-predictive identifiers: RowNumber, CustomerId, Surname
• Features include demographic, financial, and behavioural variables
• Churn rate: ~20.37%, so evaluation focuses on Recall, F1, and PR-AUC (not accuracy alone)

Leakage-Safe Pipeline (Key Design)

Preprocess → SMOTE → Model

• Preprocess: StandardScaler (numeric) + OneHotEncoder (categorical)
• SMOTE is applied inside the pipeline, so oversampling happens only on training folds (prevents leakage)
• Split: 70% train / 30% test (stratified)
• Validation: split from training only (used for threshold exploration); test set untouched until final evaluation

Models Compared

• Logistic Regression (baseline)
• Decision Tree (interpretable)
• Random Forest (300 trees)
• Gradient Boosting (best overall performance in our run)

How to Run

1) Place the dataset

Download Churn_Modelling.csv from Kaggle and place it in the same folder as churn_pipeline.py.

Note: The dataset is not included in this repository to respect Kaggle licensing.

2) Install dependencies

pip install -r requirements.txt

### 3) Run
python churn_pipeline.py