Discovering Maximal High Utility Co-location Patterns from Spatial Data

Abstract

Compared with traditional prevalent co-location patterns (PCPs), high-utility co-location patterns (HUCPs), which consider the utility of spatial features and instances, can more effectively expose interesting relationships hidden in spatial data. However, just like traditional PCPs, there are too many redundant patterns in the mining results. Users are hard to understand and apply the results. Therefore, this work proposes a concise representation of the HUCP mining results, maximal high-utility co-location patterns. Maximal HUCPs can effectively reduce redundant patterns by designing a constraint of a HUCP with its supersets. The mining results become more concentrated and convenient for users to apply. Unfortunately, the common mining methods for maximal PCPs are not suitable for discovering maximal HUCPs since the downward closure property that is often utilized to reduce the candidate search space is not available in HUCP mining. Unnecessary candidates cannot be effectively pruned in advance. Thus, a new mining algorithm, which can effectively overcome the above problem, is proposed in this work. The algorithm first enumerates all maximal cliques of the input data set. Then it arranges these maximal cliques into a special hash map structure. After that, an upper bound of the candidate maximal patterns can be determined. The instances that participate in each candidate are quickly obtained from this hash structure. Finally, maximal HUCPs are filtered efficiently. The effectiveness and efficiency of the proposed method are demonstrated through synthetics and real data sets.