Comparative analysis of association rule mining algorithms

Abstract

Data mining is one of the significant research domains in the field of computer science and it is defined as the extraction of hidden knowledge from the large data repositories. Important data mining techniques are classification, clustering, association rule generation, summarization, time series analysis and etc. Association rule is used to determine frequent patterns, association and correlations among sets of items in transactional databases, relational databases, and other information repositories. The two important steps of association rule mining are frequent item generation and association rule generation. Frequent item generation discovers all frequent sets of items; it is defined as itemsets that have at least minimum support. From these frequent items, association rules are generated. The main objective of this research is to analyze the performance of various association rule generation algorithms. The algorithms considered for this comparative analysis are Apriori, Eclat, Dclat, FP-growth, FIN, AprioriTID, Relim and H-Mine. Performance factors used are number of frequent items generated (different thresholds), memory requirements, execution time, number of rules generated (different threshold values of support and confidence) and different sizes of datasets are used. Finally from this comparative analysis we observed that the DCLAT algorithm has produced good results than other algorithms.