Reduced Search Space based Association Rule Mining Algorithm

Many algorithms have been proposed to solve the problem of mining frequent itemset. However, the large size of the itemsets search space is still a challenging problem that eliminates the performance of any association rules mining algorithm. The size of search space is exponential to the number of items in the database. In this paper, an effective mechanism is proposed to reduce the search space. Moreover, an efficient algorithm is also devised for mining association rules based on this reduced search space. An accumulative support distribution of 2-itemsets is created over different levels with only one database scan. This distribution provides estimation for the support values of all itemsets in search space. The estimated supports are used in generating the candidate itemsets in all levels without extra database scan. The proposed algorithm reduces the execution time by reducing both CPU and I/O times. CPU time is saved by reducing candidate sets size, whereas the I/O time, is reduced by reducing the required database scans. The experimental and analytical results show a significant improvement of performance up to several orders of magnitude compared to Apriori algorithm. In addition, the number of generated candidate itemsets in proposed algorithm is less than the ones generated by Apriori algorithm in most cases.