Realization of Data Mining Model for Expert Classification Using Multi-scale Spatial Data

Data mining models show great efficiency on acquiring knowledge for expert system classification. This study aimed at mining knowledge contained in landscape from multi-scale spatial data using decision tree learning model and evaluating the classification quality influenced by different scales of spatial data. Firstly, spatial data containing remote sensing images of different spatial and spectrum resolutions, digital elevation models and geographical information data with different scales were combined together to make up a spatial data infrastructure. Secondly, field samples data acquired by GPS were taken as the reference and the related spatial data was extracted. Thirdly the expert rules were developed by C5.0 decision tree models and then the rule base was used in a knowledge classifier. Finally we measured the accuracy influence of the data and data sets with different scales. The results showed: (1) The potential knowledge and rules could be detected using this data mining model with enough field samples. (2) The information provided by multi-level spatial data would influence the decision tree learning. Data set with a scale of 20m would offer most effective information. (3) After selecting effective data and scaling, we got an acceptable accuracy of 80.7% using the decision tree data mining and expert classification.