Summarizing Frequent Itemsets via Pignistic Transformation

Since the proposal of the well-known Apriori algorithm and the subsequent establishment of the area known as Frequent Itemset Mining, most of the scientific contribution of the data mining area have been focused on the study of methods that improve its efficiency and its applicability in new domains. The interest in the extraction of this sort of patterns lies in its expressiveness and syntactic simplicity. However, due to the large quantity of frequent patterns that are generally obtained, the evaluation process, necessary for obtaining useful knowledge, it is difficult to be achieved in practice. In this paper we present a formal method to summarize the whole set of mined frequent patterns into a single probability distribution in the framework of the Transferable Belief Model (TBM). The probability function is obtained applying the Pignistic Transformation on the patterns, obtaining a compact model that synthesizes the regularities present in the dataset and serves as a basis for the knowledge discovery and decision making processes. In this work, we also present a real case study by describing an application of our proposal in the field of Neuroscience. In particular, our main goal is focused on the behavioral characterization, via pignistic distribution on attentional cognitive variables, of group of children pre-diagnosed with one of the three types of ADHD (Attention Deficit Hyperactivity Disorder).