A Method of Mining Association Rules for Geographical Points of Interest

Association rule (AR) mining represents a challenge in the field of data mining. Mining ARs using traditional algorithms generates a large number of candidate rules, and even if we use binding measures such as support, reliability, and lift, there are still several rules to keep, and domain experts are needed to extract the rules of interest from the remaining rules. The focus of this paper is on whether we can directly provide rule rankings and calculate the proportional relationship between the items in the rules. To address these two questions, this paper proposes a modified FP-Growth algorithm called FP-GCID (novel FP-Growth algorithm based on Cluster IDs) to generate ARs; in addition, a new method called Mean-Product of Probabilities (MPP) is proposed to rank rules and compute the proportion of items for one rule. The experiment is divided into three phases: the DBSCAN (Density-Based Scanning Algorithm with Noise) algorithm is used to cluster the geographic interest points and map the obtained clusters into corresponding transaction data; FP-GCID is used to generate ARs, which contain cluster information; and MPP is used to choose the best rule based on the rankings. Finally, a visualization of the rules is used to validate whether the two previously stated requirements were fulfilled