Efficient Frequent Pattern Mining Using Auto-Associative Memory Neural Network

Abstract

Aims: Frequent pattern mining is one of the imperative tasks in the data mining. The soft computing techniques such as neural network, fuzzy logic have potential to be used in frequent pattern mining since these powerful tools efficiently model data which is also an essential part of mining. The proposed paper aims to provide efficient mining solution using auto-associative memory neural network to efficiently traverse and reduce the search space, and to reduce the I/O computations. It also aims to keep balance in computational and resource efficiency. Methodology: This paper proposes an efficient algorithm for mining frequent patterns using auto-associative memory. Auto-associative memory is best suitable artificial neural network (ANN) approach for association rule mining as it stores associations among the patterns. In the proposed system auto-associative memory based on the correlation matrix memory (CMM) is used to find the frequent patterns. The proposed work introduces novel learning and recall algorithms using CMM for mining frequent patterns efficiently. The proposed learning algorithm reduces the search space tremendously for recall mechanism. The proposed recall algorithm uses only frequent 1-patterns and frequent 2-patterns for determining all other frequent patterns, reducing the number of I/O computations and speeding up the mining process. This approach keeps well balance among computational and resource efficiency. Conclusion: The performance of the proposed system is compared with traditional algorithms like Apriori, Frequent pattern growth (FP-growth), Compressed FP-tree based algorithm (CT-PRO) and Linear time Closed itemset Miner (LCM). The experimental results show order of magnitude improvement in execution time and storage space optimization to accumulate frequent patterns. Proposed work proves milestone approach in the field of frequent pattern mining using artificial neural network.