ASSESSMENT OF CLASSIFIERS THROUGH DECISION TREE AND REGRESSION TREE ALGORITHMS IN URBAN AREA USING WORLDVIEW-2 IMAGE

Geographic Object-Based Image Analysis allows the simulation from the view of a human interpreter using knowledge models expressed by semantic networks. Data mining techniques have been widely used as a support tool for the construction of the semantic network. In this sense, the aim of this study is to analyze the performance of the CART and C4.5 algorithms, which use decision trees, to classify urban land cover. A WorldView-2 image was used for this analysis. Both algorithms presented good accuracy. The C4.5 algorithm accuracy presented average values slightly higher than the CART algorithm. C4.5 was supported by other software for the execution of the analyses. This posed a challenge to the researchers for data integration, data format conversion and also file replication. Differently, the CART algorithm tested is part of an integrated GEOBIA platform, which benefits the user reducing the time spent to execute all the image analysis steps.