Case Study · Machine Learning

Customer Churn Prediction Model

A machine learning project predicting customer churn using classification techniques: logistic regression and ensemble methods reaching 85%+ accuracy, with XGBoost and feature engineering driving a ~12% improvement over the baseline. A full analytical report with visualizations supports data-driven retention decisions.

October 2025 – November 2025 PythonXGBoostScikit-learnPandas
01

results

~73%
BASELINE ACCURACY
85%+
FINAL ACCURACY
+12%
LIFT VIA FEATURE ENGINEERING + XGBOOST
ROC-AUC
PRIMARY EVALUATION METRIC
02

pipeline

Every customer passes through the ensemble model once. Most flow through to "Retained," and a smaller share gets flagged as "Churned."

Customer Pool
XGBoost + LogReg
85%+ accuracy
Retained
Churned
03

features

Ensemble Methods

XGBoost classifier combined with logistic regression, boosting prediction performance ~12% above the baseline model.

Model Evaluation

ROC-AUC scoring, confusion matrices, and precision/recall analysis to validate and compare model performance.

Analytical Report

Detailed report with key statistics, visualizations, and insights supporting data-driven retention strategies.

04

stack

Python 3Scikit-learnXGBoostPandasNumPyMatplotlibSeabornROC-AUC
05

signals

Tenure, contract type, monthly charges, and support call volume were the strongest predictors of churn identified in feature importance analysis.