Gradient Boosting
Gradient Boosting delivered the best results, achieving an R² of 0.5475 and RMSE of 2.594. The model effectively captured nonlinear relationships and generalized well to unseen data, making it ideal for forecasting Urban Heat Island Intensity (UHII). The best parameter for the model were: Learning rate =.1, max_depth =4 and number of trees = 150.
Random Forest
Random Forest produced moderate performance with an R² of 0.5531 and RMSE of 2.579. While it showed good generalization, its lack of spatial reasoning limited its ability to capture localized heat patterns. The best parameters were: Learning rate = .1, max_depth = 4 and number of trees = 16
Geographically Weighted Regression
Geographically Weighted Regression achieved an R² of 0.65 and RMSE of 2.29, offering valuable spatial insights by modeling local variations in UHII. It was particularly effective in rural and mountainous areas where spatial heterogeneity is prominent.