Customer Churn Prediction Model

Business Problem: The company experienced a rising churn rate but lacked a reliable way to identify at-risk customers early. Existing retention strategies were reactive, generic, and costly, resulting in lost revenue and inefficient marketing spend. Role: Data Analyst & Machine Learning Practitioner responsible for data preparation, feature engineering, model development, evaluation, and insight communication. Data Sources: • Customer demographics • Transaction history • Engagement metrics (logins, activity frequency) • Support interactions • Subscription lifecycle data Tools & Technologies: Python, Pandas, NumPy, Scikit-Learn, Matplotlib, Seaborn, Jupyter Notebook Process: 1. Data Collection - Imported structured customer datasets from SQL and CSV files. - Merged multiple tables into a unified analytical dataset. 2. Data Cleaning - Handled missing values using median and mode imputation. - Removed duplicate customer records. - Normalized skewed numerical features. 3. Feature Engineering - Created new variables such as: • Tenure length • Average monthly spend • Days since last login • Support ticket frequency - Encoded categorical variables using one-hot encoding. - Scaled numerical features for model stability. 4. Modeling - Tested multiple algorithms: Logistic Regression, Random Forest, XGBoost. - Selected Random Forest for best balance of accuracy and interpretability. - Tuned hyperparameters using GridSearchCV. 5. Evaluation - Achieved: • 86% accuracy • 0.91 precision for identifying churners • 0.88 recall - Used confusion matrix, ROC curve, and feature importance analysis. Key Insights: • Low engagement in the first 30 days was the strongest predictor of churn. • Customers with 3+ support tickets in a month were 2.4x more likely to churn. • High-value customers churned less frequently but required targeted onboarding. • Tenure under 90 days showed the highest churn probability. Business Impact: • Enabled proactive retention campaigns targeting high-risk customers. • Reduced churn by an estimated 12% in pilot testing. • Improved marketing efficiency by focusing incentives on the right segments. • Provided leadership with a predictive framework for long-term customer strategy.