Comparing three methods for modeling the uncertainty in knowledge discovery from area-class soil maps

Knowledge discovery has been demonstrated as an effective approach to extracting knowledge from existing data sources for soil classification and mapping. Soils are spatial entities with fuzzy boundaries. Our study focuses on the uncertainty associated with class assignments when classifying such entities. We first present a framework of knowledge representation for categorizing spatial entities with fuzzy boundaries. Three knowledge discovery methods are discussed next for extracting knowledge from data sources. The methods were designed to maintain information for modeling the uncertainties associated with class assignments when using the extracted knowledge for classification. In a case study of knowledge discovery from an area-class soil map, all three methods were able to extract knowledge embedded in the map to classify soils at accuracies comparable to that of the original map. The methods were also able to capture membership gradations and helped to identify transitional zones and areas of potential problems on the source map when measures of uncertainties were mapped. Among the three methods compared, a fuzzy decision tree approach demonstrated the best performance in modeling the transitions between soil prototypes.