Data Mining Association Rules using Probabilistic Functions without Predefined Weights

Abstract

Data mining is the procedure of identifying the important and relevant data from large heterogeneous databases. Data mining plays an important role because of its usage in various domains. The transaction in the data mining defines the profit of the items associated with it. Earlier algorithms were proposed to measure the w-support without assigning predefined weights to determine the important transactions using the HITS model. Significant items are extracted from the databases using the quality of the transactions. However, there is considerable overhead in computing the w-support, as it requires four to five iterations. In this paper, two algorithms are proposed which uses the Poisson distribution and Normal distribution while computing the w-support without using the pre-assigned weights. The Poisson distribution uses the probability mass functions whereas the Normal distribution uses the probability density function to compute the w-support. The experiments were executed on various standard datasets. The results of our proposed algorithms show a considerable decrease in normalization time to compute the w-support as compared to the HITS model. Hence our algorithms provide better performance with respect to execution time and a number of significant items.