Off-grid solar systems encounter difficulties during low solar irradiance periods, particularly at night when photovoltaic generation stops. This study evaluates and compares the performance of five machine learning models Logistic Regression, Random Forest, Gradient Boosting, Support Vector Machine (RBF), and Decision Tree for prioritizing load cut-offs in AI to maintain critical loads during energy shortages. Data averaged hourly from an off-grid solar setup in Chiang Mai, Thailand, for 2024, included photovoltaic output, battery metrics, and categorized load usage during rainy, winter, and summer seasons. Models were trained on rainy season data and tested on winter, summer, and October datasets. We evaluated performance through MAE, RMSE, R², classification reports, F1-scores, and k-fold cross-validation to ensure stability. Results indicate that Random Forest and Gradient Boosting consistently reached the highest accuracy (R² > 0.95 in most seasons) with low MAE and RMSE, whereas Decision Tree and Logistic Regression showed more variability. AI-driven scenarios greatly improved nighttime battery performance over non-AI approaches, especially during the rainy season. This method enhances energy reliability and battery longevity in off-grid settings, but outcomes vary by location. Future research should explore a wider range of climates and load profiles.