An Effective Approach on Mining Co-Location Patterns from Spatial Databases with Rare Features

A co-location pattern is a group of spatial features whose instances are frequently appearing together in geography. Co-location pattern mining is particularly valuable for discovering spatial dependencies. Lots of co-location pattern mining approaches have been proposed, but they often emphasize the equal participation of every spatial feature. As a result, the interesting pattern which involves spatial features with significantly different for the number of instances cannot be captured. In this paper, we are committed to address the problem of mining co-location patterns from the spatial database with rare features. Specifically, we first propose a new interest measure, namely the weighted participation index. This interest measure is related to the distribution of the number of instances for spatial features, and it has ability to capture the prevalent co-location patterns with or without rare features. Furthermore, we prove that the weighted participation index possesses the approximate monotonicity property, which can be utilized to improve the computational efficiency, and thereby an efficient algorithm is developed. As demonstrated by extensive experiments, our approach is effective, efficient and scalable for mining co-location patterns embedded in the spatial database with rare features.