Comparison of Candidate Itemset Generation and Non Candidate Itemset Generation Algorithms in Mining Frequent Patterns

Abstract

Association rule mining is one of the important techniques of data mining used for exploring fruitful patterns from huge collection of data. Generally, the finding of frequent itemsets is the most significant step in association rules mining, and most of the research will be centered on it. Numerous algorithms have been discovered to find effective frequent itemsets. This paper compares the frequent pattern mining algorithms that use candidate itemset generation and the algorithms without candidate itemset generation. In order to have on field simulation for comparison, a case study algorithm from both types was chosen such as ECLAT and FP-growth algorithms. Equivalence class clustering and bottom up lattice traversal (ECLAT) algorithm accommodates ?Depth First Search? approach and requires the generation of candidate itemset. The FP-growth algorithm follows the ?Divide and Conquer? method and does not require candidate itemset generation. In this paper, the benchmark databases considered for comparison are Breast Cancer, Customer Data, and German Data etc. The performances of both the algorithms have been experimentally evaluated in terms of runtime and memory usage. From the result it is analyzed that the FP-tree algorithm is more advantageous as it does away with the need of generation of candidate patterns.