isp-churn-prediction

A machine learning classification project that helps an Internet Service Provider in keeping existing subscribers by predicting if an existing customer will churn or not.

I highly recommend you to check out the slides for getting an overview of the project.

Data Sources

Internet Service Provider Customer Churn Dataset by Mehmet Sabri Kunt on Kaggle.

Problem Type

Binary Classification of predicting customer churn.

Evaluation Metric

ROC-AUC

Comparative Analysis of Models

Algorithm	Mean AUC-ROC	Standard Deviation
Logistic Regression	0.9624	0.0027
K-Nearest Neighbors	0.9748	0.0020
Decision Tree	0.9751	0.0021
Random Forest	0.9827	0.0015
XGBoost	0.9840	0.0014

Results produced by Final Model(XGBoost Classifier)

Confusion Matrix

ROC-AUC

Train Set: 0.9979611536456878
Test Set: 0.9842082218633272

f1-Score

Train Set: 97.95469985401037
Test Set: 95.16816245127367

Accuracy

Train Set: 97.72774717774959
Test Set: 94.67

Tool, Libraries & Frameworks used

• Numpy
• Pandas
• Matplotlib
• Scikit-Learn
• Seaborn
• XGBoost
• Optuna
• DVC
• AWS