Clustering transactions based on weighting maximal frequent itemsets

We propose a new similarity measure for comparing maximal frequent itemset (MFI), which takes into account not only non-numeric attributes but also numeric attributes of each item while computing similarity between MFIs. This provides more reliable soundness for clustering results interpretation. Traditional approaches consider just one side and are apt to lead to unintelligible clustering results for decision-makers. Based on properties of maximal frequent itemset (MFI), we construct a multi-level hierarchical model (MHM) for our clustering algorithm. Moreover, to evaluate our approach and compare with other similarity strategies, we construct an original evaluating strategy NF_Measure which integrates both quantity similarity and quality similarity between transactions. We experimentally evaluate the proposed approach and demonstrate that our algorithm is promising and effective.