FP-growth 방법을 이용한 시간 가중치 Fuzzy 연관 규칙 마이닝

Abstract

Data mining is the process of extracting meaningful knowledge or interesting patterns from existing databases for specific purposes. Normally, real transaction data consists of quantitative values and a delicate data mining algorithm should handle this kind of data. The fuzzy concepts are used in the mining algorithm for discovering useful fuzzy association rules from quantitative data. Previous Fuzzy Apriori approaches in this field need enormous amount of calculation to find multiple frequent pattern from multiple candidate itemsets because of fuzzy count. In this paper, we propose a fuzzy association rule mining method using FP-growth method which significantly reduces candidate itemsets to save the run-time and memory. This method uses compressed the database representing frequent items into a frequent-pattern tree which retains the itemset association information. And we adopt a time weighted method for partitioning transactions in the database using exponential smoothing. Transactions are assigned different weights as time goes on. This implies that our method helps to reflect time tendency to the transaction database and extract more meaningful fuzzy association rules. Our performance study shows that FARFP method is more efficient than Fuzzy Apriori algorithm handling large scale of multidimensional database.