Customer Churn Prediction with Explainable AI

A production-ready, modular ML pipeline designed to solve the 'black-box' problem in predictive modeling. The system implements a clean separation of concerns—from data ingestion and specialized preprocessing to evaluation and model explanation. By integrating SHAP (SHapley Additive exPlanations), the project provides granular insights into which customer behaviors (e.g., tenure, billing charges) most significantly impact the model's predictions, enabling data-driven retention strategies.

Key Features

• Modular Architecture: Decoupled Python components for data loading, preprocessing, training, and evaluation for high maintainability.
• Explainable AI (XAI) Integration: Leveraged SHAP to generate summary plots that visualize feature importance and local model decisions.
• Robust Preprocessing: Automated handling of dirty data, including type coercion for numerical fields and median imputation for missing values.
• Comprehensive Evaluation: Performance tracking using ROC-AUC scores and detailed classification reports (Precision/Recall/F1-Score).
• Automated Insight Generation: System automatically exports visual interpretability reports for non-technical stakeholders.
• Feature Engineering: Implemented efficient One-Hot Encoding and feature alignment for Random Forest compatibility.

Tech Stack

Screenshots