Forecasting Experiment Toolkit

Anonymized professional experimental work on tabular accident-data preprocessing, geospatial features, chronological validation, and model comparison/training utilities. Contribution evidence includes training progress/checkpoint/resume work, candidate configuration, historical-rate/hurdle approaches, and controlled-comparison tooling. Project source includes scikit-learn/PyTorch and H3-related workflows. Scope: Research/experiments, not verified production forecasting. Leakage risks, comparable data splits, and evaluation methodology remain essential; no accuracy or operational-improvement figure is asserted.