C4.5 Decision Tree Algorithm for Spatial Data, Alternatives and Performances

Using data mining techniques on spatial data is more complex than on classical data. To be able to extract useful patterns, the spatial data mining algorithms must deal with the representation of data as stack of thematic layers and consider, in addition to the object of interest itself, its neighbors linked through implicit spatial relations. The application of the classification by decision trees combined with the visualization tools represents a convenient decision support tool for spatial data analysis. The purpose of this paper is to provide and evaluate an alternative spatial classification algorithm that supports the thematic-layered data organization, by the adaptation of the C4.5 decision tree algorithm to spatial data, named S-C4.5, inspired by the SCART and spatial ID3 algorithms and the adoption of the Spatial Join Index. Our work concerns both data organization and the algorithm adaptation. Decision tree construction was experimented on traffic accident dataset and benchmarked on both computation time and memory consumption according to different experimentations: study of phenomenon by a single and then by multiple other phenomena, including one or more spatial relations. Different approaches used show compromised and balanced results between memory usage and computation time.