churn-prediction-lgbm

ChurnGuard — E-Commerce Customer Churn Prediction (LightGBM)

A LightGBM-based model for predicting customer churn in e-commerce, trained on the Kaggle E-Commerce Churn Dataset.

Model Details

• Model: LightGBM (via imbalanced-learn pipeline with SMOTE)
• Task: Binary classification (Churn: 0 = Stayed, 1 = Churned)
• Dataset: 5,630 customers x 30 features (after cleaning & feature engineering)
• Training: Stratified 5-fold cross-validation with SMOTE on training folds only

Performance

Files

Usage

import joblib
import pandas as pd

# Load model and feature names
model = joblib.load("best_model.joblib")
feature_names = joblib.load("feature_names.joblib")

# Predict on new data
sample = pd.DataFrame([{
    "Tenure": 4, "CityTier": 3, "WarehouseToHome": 6,
    "Gender": 1, "HourSpendOnApp": 3, "NumberOfDeviceRegistered": 3,
    "SatisfactionScore": 2, "MaritalStatus": 0, "NumberOfAddress": 9,
    "Complain": 1, "OrderAmountHikeFromlastYear": 11,
    "CouponUsed": 1, "OrderCount": 1, "DaySinceLastOrder": 5,
    "CashbackAmount": 159.93, "tenure_bucket": 0,
    "engagement_score": 3, "cashback_per_order": 159.93,
    "is_recent_buyer": 0, "has_multi_device": 0, "is_high_spender": 0,
    "PreferredLoginDevice_Mobile Phone": 1,
    "PreferredPaymentMode_Credit Card": 0,
    "PreferredPaymentMode_Debit Card": 1,
    "PreferredPaymentMode_E wallet": 0,
    "PreferredPaymentMode_UPI": 0,
    "PreferedOrderCat_Grocery": 0,
    "PreferedOrderCat_Laptop & Accessory": 1,
    "PreferedOrderCat_Mobile Phone": 0,
    "PreferedOrderCat_Others": 0,
}])

prediction = model.predict(sample)
probability = model.predict_proba(sample)[:, 1]

print(f"Churn prediction: {prediction[0]}")
print(f"Churn probability: {probability[0]:.4f}")

SHAP Explainability

import shap
import joblib

explainer = joblib.load("shap_explainer.joblib")
shap_values = explainer.shap_values(sample)
shap.waterfall_plot(shap.Explanation(
    values=shap_values[0],
    base_values=explainer.expected_value,
    data=sample.iloc[0],
    feature_names=feature_names,
))

Project

Part of the ChurnGuard platform — GitHub Repository

Data Exploration Project | ITI Alexandria | Track AI | INTAKE 46