Mining Outliers in Spatial Networks

Outlier analysis is an important task in data mining and has attracted much attention in both research and applications. Previous work on outlier detection involves different types of databases such as spatial databases, time series databases, biomedical databases, etc. However, few of the existing studies have considered spatial networks where points reside on every edge. In this paper, we study the interesting problem of distance-based outliers in spatial networks. We propose an efficient mining method which partitions each edge of a spatial network into a set of length d segments, then quickly identifies the outliers in the remaining edges after pruning those unnecessary edges which cannot contain outliers. We also present algorithms that can be applied when the spatial network is updating points or the input parameters of outlier measures are changed. The experimental results verify the scalability and efficiency of our proposed methods.