Wildlife Hotspot Detector
Engineered feature vectors (HOG, GLCM) to isolate texture variance, distinguishing signal from background Quantified feature importance via Random Forest ranking, reducing 750+ dimensions to maximize signal fidelity Architected Active Learning loops via Uncertainty Sampling, utilizing SMOTE to resolve class imbalance Optimized XGBoost hyperparameters via GridSearchCV, achieving 0.85 F1 on highly skewed datasets