Telco Customer Churn Analysis

Customer churn is one of the most expensive problems a subscription business faces — and one of the most preventable. This project takes a telecom dataset of 7,032 customers and builds a complete analytical pipeline from raw data to an operationally deployable retention system.

The analysis moves through five stages. The first establishes a clean baseline and confirms a 26.5% churn rate. The second uses targeted visualizations to surface four key patterns: a 42.7% churn rate among month-to-month contract holders, a concentration of churners in the first 12 months of tenure, a counterintuitive churn spike among Fiber Optic customers despite their premium pricing, and a near 3× difference in churn between customers with and without tech support. The third stage validates every one of these patterns statistically, Chi-Square tests for categorical variables and Mann-Whitney U tests for numerical ones, with test selection justified by a formal skewness check rather than assumed.

The fourth stage trains and compares four classification models, Logistic Regression, Decision Tree, Random Forest, and Gradient Boosting evaluated on recall rather than accuracy, given the class imbalance. The fifth stage is where the project moves beyond standard practice: cohort analysis tracks how different joining periods behave over time, Kaplan-Meier survival curves with log-rank testing quantify how long different customer segments typically last, and a CLV risk segmentation matrix crosses projected customer lifetime value against churn probability to identify a specific CRITICAL segment, high-value customers likely to leave with a total revenue figure attached.

The final output is a two-dashboard Tableau workbook framed around two questions a business actually asks: where is churn happening, and why. Every finding connects directly to a recommended action.

Tools Used: Python - pandas - scikit-learn - lifelines - Tableau Public - Statistical Testing - Survival Analysis

Key skills demonstrated: Feature engineering, classification modeling, cross-validation, statistical testing, dashboard design, business storytelling