AI-Driven Post-Wildfire Ecosystem Recovery
Won a competitive $2,500 grant to build a machine learning framework for vegetation recovery and reforestation planning across the Thompson-Okanagan. Merged six environmental datasets into a reproducible 2,600 record master set with 32 engineered features over six BEC zones, benchmarked five classifiers with SMOTE oversampling and Optuna tuning, and landed on LightGBM at 0.707 weighted F1. SHAP named burn severity the dominant predictor. Shipped a native species recommender, wrote the paper in LaTeX, and presented it at the TRU Sustainability Conference.
TRU Student Sustainability Research Grant ($2,500) · Supervised by Dr. Ghazanfar Latif
