jaideep005/churn_retention_system

ChurnGuard AI is a web app that predicts which customers are likely to stop using a service and then suggests personalized strategies to keep them. It is built for businesses that want to identify at-risk customers before they leave, rather than reacting after the fact. The app works through five steps shown as tabs in the browser. First, it loads customer data, either from a public IBM telecom dataset or from a custom API endpoint you provide. Second, it cleans and prepares the data automatically, calculating things like estimated customer lifetime value, engagement scores, and risk segments. Third, it trains two machine learning models (Logistic Regression and Random Forest) on that data, runs cross-validation to measure accuracy, and picks the better-performing one to generate predictions for all customers. Fourth, it displays a dashboard with charts showing churn probability distribution, revenue at risk, and how customers break into low, medium, and high risk groups. Fifth, it generates a retention strategy for each at-risk customer using an AI language model. For the retention strategies, the app supports three modes. If you do not have an API key, it uses a built-in rule-based template that produces reasonably detailed suggestions. If you have a Google Gemini or OpenAI API key, it uses that model instead to produce more tailored recommendations. Once analysis is complete, you can export a CSV file that includes each customer's churn probability, predicted outcome, risk segment, and estimated 24-month lifetime value. The file is formatted for easy import into Power BI dashboards. The app runs locally via Streamlit and opens in a browser at localhost. It requires Python 3.9 or higher and a standard set of data science libraries. No database or server setup is needed beyond installing the dependencies.