End-to-End Telco Customer Churn Prediction System

Developed and deployed a production-grade Machine Learning web application designed to predict telecom customer churn and identify key drivers of customer attrition. This project bridges the gap between data science research and real-world deployment by transforming a Jupyter Notebook workflow into an interactive, cloud-hosted web service. Key Technical Highlights & Workflow: Exploratory Data Analysis & Preprocessing: Conducted thorough EDA on customer behavioral data, handled missing values, encoded categorical features, and performed feature scaling to ensure robust model performance. Model Training & Hyperparameter Tuning: Trained multiple machine learning algorithms (including Random Forest, XGBoost, and Logistic Regression) and optimized hyperparameters to achieve high predictive accuracy and generalization. Model Explainability (SHAP): Integrated SHAP (SHapley Additive exPlanations) values to make complex model predictions transparent, helping stakeholders easily interpret feature impacts (such as tenure, monthly charges, and contract types) on individual predictions. Backend & API Architecture: Built a high-performance RESTful API using FastAPI to handle real-time inference requests seamlessly, coupled with a clean, responsive frontend user interface. Cloud Deployment & DevOps: Successfully packaged the application and trained model (.pkl), navigated serverless deployment constraints by leveraging container-based cloud infrastructure (Render), and ensured continuous availability. Tech Stack: Python, Scikit-Learn, XGBoost, Pandas, NumPy, FastAPI, Uvicorn, SHAP, HTML/CSS, Git, Render.