Mining Association Rules Based on Deep Pruning Strategies

Today mobile network and various smart devices flourish rapidly. Data collected from the mobile devices and network can bring us huge opportunities to understand some significant characteristics of the users which traditional data cannot. Association rules mining is an extremely important topic in data mining that can make the utmost value of massive data effectively. Apriori algorithm and the improved Apriori ones based on Boolean matrix are the representative ones in association rules mining. Nevertheless, these solutions have their problems. In this paper, we have proposed an algorithm called MAR-DPS, which has some deep pruning strategies containing three methods to compress the size of frequent itemsets and reduce the joining numbers in generating new frequent itemsets. It can also select the appropriate method to generate frequent 2-itemsets when facing different data sets. Extensive experimental results on three different data sets have demonstrated that our MAR-DPS can perform much better than other tested algorithms.