This repository provides a reproducible workflow for predicting soil erosion rates using multiple machine learning models (convolutional (CNN), deep (DNN), single-hidden-layer (SNN) neural network, and Random Forest(RF)). It includes data preprocessing, model training with cross‑validation, evaluation metrics, and inference to generate GeoTIFF prediction maps. V.2: Updated to fully use leave-on-area-out (LOAO) approach. Updated new results and added grid search pipeline.