An application of spatial decision tree for classification of air pollution index

Abstract

A decision tree is an analysis skill and a classification algorithm, whose basic principle is the combination of probability theory and an analysis tool of tree shapes. It derives a hierarchy of partition rules with respect to a target attribute of a large dataset. Nowadays, concrete coordinates exist in lots of datasets, which leads to the spatial distribution of datasets. However, conventional decision tree does not take the spatial distribution of records in the dataset into account, which makes it inadequate to deal with the geographical datasets. A number of new approaches to the analysis of geographical data have been proposed in recent years. In the purpose of evaluating the application of a spatial entropy-based decision tree, a spatial entropy-based decision tree that employed to classify the air pollution index (API) is presented in this paper. A spatial decision tree differs from a conventional tree in the way that it considers the spatial autocorrelation phenomena in the classification process. At each level of a spatial decision tree, the supporting attribute that gives the maximum spatial information gain is selected as a node. A case study oriented to the classification of API, whose study area is main cities in China, deals with the norms of the API, including density of total suspended particulate, density of SO2, density of NO2, and etc. After the process of data processing, and graphical analysis, it demonstrates a tree shape of the classification of the API and a map of the spatial distribution of the target attribute's categories, which illustrate the practicability of spatial decision tree.