Miuul Capstone project integrating multi-source data — USGS seismic records, NASA meteorological data, and astronomical variables — into an end-to-end ML pipeline that predicts risk probabilities and estimated timelines for major seismic events, served through an interactive Streamlit app.
Engineered features from 79 explanatory variables and handled extensive missing data and skew in the Ames Housing dataset. Fine-tuned an ensemble of Random Forest and XGBoost, reaching an RMSE of 0.118 on log-transformed sale prices.
Processed and engineered features from a Scoutium dataset of 10,700+ scouted football player observations. Trained a classifier reaching 85% accuracy to sort players into "average" and "highlighted" segments for data-driven recruitment.
Designed and deployed an XGBoost churn model reaching an 87% F1-score, identifying key at-risk customer segments, and served through a Streamlit app wired to a REST API for real-time predictions.