An Approximate Approach to Frequent Itemset Mining

Eclat algorithm is one of the most widely used frequent itemset mining methods. One significant bottleneck of the Eclat algorithm is that the efficiency for calculating the intersection of itemsets is low especially when the itemsets have a large number of transactions. In this work, for the purpose of efficiency improvement and resource saving, we propose an approximate variation of Eclat algorithm which is based on MinHash technique. A recent research shows that MinHash technique can be used for estimating the size of the intersection of sets. So we investigate the role of MinHash technique to overcome the drawback which inefficiency of itemsets intersection in Eclat algorithm. This approach can provide an approximate answer which is more useful than 'exact' result in many situations. In this work, we propose an approximate algorithm called HashEclat. Our algorithm considers the tradeoff between accuracy of the mining results and algorithm execution time. Both of theoretical analysis and realistic experiments show that it can output almost all the frequent itemsets with faster speed and less memory space. Besides other optimization methods of the Eclat algorithm can also be used in our algorithm, since our method only accelerates calculating the intersection speed.