This project builds a complete ML pipeline to model medical insurance costs based on demographic and lifestyle features. The workflow includes exploratory data analysis (EDA), preprocessing, multiple regression models, hyperparameter tuning, ensemble stacking, log-transform modeling, SHAP interpretability, and full residual diagnostics.

The online payment fraud analysis project follows several step approach from data preprocessing through model evaluation, result comparison and final model selection, using transaction patterns to identify fraud indicators including account draining, suspicious transfers, and balance inconsistencies.

A machine learning web app that predicts a student’s expected score and grade using self-study hours, attendance, and class participation, also showing performance insights such as risk level, focus area, and improvement recommendations. It is built with React, FastAPI, and scikit-learn. Deployment: Vercel for frontend, Render for backend.

Leakage-aware flood susceptibility mapping using classical machine learning and strict spatial holdout, with reproducible preprocessing, explainability analysis, and benchmark comparison across eight models.

Binary classification to predict employee departures using HR, survey & evaluation data — HistGradientBoosting with SMOTE resampling, SHAP interpretability, 75% recall on test set

Stroke prediction using machine learning (LogReg, RF, GBoost, LightGBM) with class imbalance handling and threshold optimisation

Trees Fungal Disease Predictive Model / Penn State University

DS Take Home Assessment