Customer Churn Prediction App

Project Overview #

This project tackles the challenge of predicting customer churn using machine learning. I built and compared three models — Logistic Regression, Random Forest, and a tuned Random Forest — to identify which customers are most likely to leave a service.

The models were trained on telecom customer data, and I deployed all three in an interactive Streamlit app. Users can explore each model’s performance and test different service scenarios to see how likely a customer is to churn.

Business Value #

Customer churn is a major cost driver in subscription-based services. By predicting churn in advance, companies can offer proactive support, targeted incentives, or long-term contracts to reduce attrition.

This app helps businesses:

• Identify high-risk customers based on usage and contract patterns
• Visualize key drivers of churn
• Make data-backed retention decisions

Key Results #

The Logistic model caught more churners (high recall), while the Random Forest was better at avoiding false positives (higher precision). Both models had solid ROC-AUC scores, and tuning helped balance Random Forest performance.

Insights from the Data #

• Customers with high charges and short tenure are most likely to churn.
• Those on two-year contracts are far more likely to stay.
• Fiber optic service shows a stronger link to churn — possibly due to pricing or performance concerns.

Tools Used #

• Python (Pandas, NumPy)
• Scikit-learn (modeling + evaluation)
• Streamlit (app deployment)
• Matplotlib & Seaborn (visualizations)
• GridSearchCV (model tuning)

Try It Yourself #

• Launch the App
• View the Code on GitHub

App Interface Preview #

Here’s what the deployed app looks like in action — with real-time predictions and model explainability:

What I Learned #

This project strengthened my skills in:

• Working with imbalanced datasets using class weights
• Comparing models not just on accuracy but precision and recall
• Deploying ML projects with Streamlit for real-time interaction