Abstract Accurate spatial prediction of shallow foundation bearing capacity remains challenging in geologically complex terrains where conventional interpolation methods fail to adequately capture abrupt geological transitions. This study introduces an integrated KNN-GIS framework that combines predictive modeling with quantitative validation and systematic bias correction. Using data of 410 boreholes in the Pokhara Valley, Nepal, an optimized KNN model was developed and validated through comprehensive stratum-wise geological analysis. The framework establishes a validation system that categorizes geological formations into four classes (A-D) based on predictive accuracy, identifies systematic biases in karst-prone formations, and provides mathematically derived correction factors to address these limitations. A three-phase engineering decision system translates validation results into actionable investigation protocols, producing spatially explicit bearing capacity maps with quantified uncertainty bounds. The methodology successfully delineates high-risk karst zones from stable bedrock areas while offering a clear and reproducible workflow for preliminary site assessment. This integrated approach bridges machine learning predictions with engineering reliability, providing a robust, transferable tool for risk-informed urban development in geologically challenging environments worldwide.
Karki et al. (Thu,) studied this question.