PREDICTION OF SOIL PARTICLES USING A SPATIALLY ADAPTIVE GEOGRAPHICALLY WEIGHTED K-NEAREST NEIGHBORS ORDINARY LOGISTIC REGRESSION APPROACH

Soil particle prediction is crucial in various fields, including agriculture, environmental management, and geotechnical applications. The spatial variation of soil texture significantly affects land fertility, erosion risk, and construction feasibility. However, conventional statistical methods and machine learning techniques often fail to capture the complex spatial heterogeneity in soil distribution. This study proposes the Geographically Weighted K Nearest Neighbors Ordinary Logistic Regression (GWKNNOLR) method to improve the accuracy of soil particle classification by integrating geographically weighted regression with an adaptive spatial weighting mechanism using the K Nearest Neighbors (KNN) algorithm. The objective of this research is to develop and evaluate a spatially adaptive classification model that more accurately predicts soil particle categories, namely sand, silt, and clay, by incorporating local spatial dependencies using GWKNNOLR in the Kalikonto watershed (DAS Kalikonto) in Batu. This study utilizes field measurement data combined with digital terrain modeling to analyze the relationship between local morphological variables and soil texture classification (sand, silt, and clay). The study area includes 50 observation points and 8 test variables. The model's performance is compared to the Ordinary Logistic Regression (OLR) method. The results indicate that GWKNNOLR achieves a classification accuracy of 88 percent, outperforming OLR, which only reaches 80 percent. Integrating KNN as a spatial weighting mechanism enhances adaptability to variations in sample distribution, leading to more accurate predictions. These findings emphasize the importance of considering spatial dependencies in soil texture modeling. The proposed method can support sustainable land resource management, erosion risk mitigation, and precision agriculture by providing more reliable soil classification. Future research may explore further optimization of spatial weighting mechanisms and the application of this method in different geographical regions.