Spatial Data Mining for Prediction of Unobserved Zinc Pollutants Using Various Kriging Methods

Following years of contamination, rivers may experience sig­nificant levels of heavy metal pollution. Our research aims to pinpoint hazardous areas in these rivers. In our specific case, we focus on the floodplains of the Meuse River contaminated with zinc (Zn). Elevated zinc concentrations can lead to various health issues, including anemia, rashes, vomiting, and stomach cramping. However, due to limited sample data on zinc con­centrations in the Meuse River, it becomes imperative to gen­erate missing data in unidentified regions. This study employs universal Kriging in spatial data mining to investigate and pre­dict unknown zinc pollutants. The semivariogram serves as a valuable tool for illustrating the variability pattern of zinc. To predict concentrations in unknown regions, the model captured is interpolated using the Kriging method. Employing regression with geographic weighting allows us to observe how stimu­lus-response relationships change spatially. Various semivario­gram models, such as Matern, exponential, and linear, are uti­lized in our work. Additionally, we introduce Universal Kriging and geographically weighted regression. Experimental findings indicate that: (i) the Matern model, determined by calculating the minimum error sum of squares, is the most suitable theoret­ical semivariogram model; and (ii) the accuracy of predictions is visually demonstrated by projecting results onto a real map.

کلیدواژه ها:

spatial data mining ، missing data ، semivariogram ، Universal Kriging ، Geographically weighted ، Regression