Ahmed Saad

Predictive Maintenance for OBD-II Engine Faults — graduation capstone, 5-person team (project lead) • Led the project end to end — scoping, data collection, modeling, and validation — for an OBD-II predictivemaintenance system: data pipeline, XGBoost fault classifier, and live dashboard. • Developed the Android (Kotlin) data-collection app reading 24 engine PIDs over Bluetooth (ELM327); ran on-vehicle testing and validation on a real car. • Redesigned model validation from leave-one-scenario-out to leave-one-family-out cross-validation to match the deployment scenario, raising worst-fold macro-F1 from ~0.81 to ~0.96. Cervical Vertebra Detection from CT-Scan Metadata — RSNA 2022 dataset • Built multi-label classifiers predicting cervical vertebra visibility (C1–C7) from 711K rows of CT-slice metadata; best model (XGBoost) averaged 0.932 F1 and 0.997 ROC-AUC across seven targets. • Prevented patient-level data leakage with group-aware train/test splitting across 2,019 CT studies; benchmarked four algorithms with GridSearchCV hyperparameter tuning