Performance Comparison of FPCFI and ReLIM Methods for Data Stream Mining with Sliding Window Preprocess

Abstract

Recently, data stream mining has been widely studied for its increasing application potentials. Data stream is an example of data which is real-time, continuous, ordered sequence of items with various update rates and is potentially unlimited in the amount of data. Data stream mining is important for applications that require complete frequency information as well as high time and space efficiencies. Because of the characteristics of data stream which have potentially unlimited amount of data, algorithms which give good performance and results are required. This research uses sliding window technique for preprocessing data stream. The data stream mining techniques which are used for extracting patterns between generated itemsets are FPCFI and ReLIM. The patterns to be mined are the closed frequent itemsets. The performances of the algorithms will be compared using ANOVA.The data used in this research includes Webdocs and Accidents. FPCFI required a higher time usage i.e. 0.47s, 0.78s, 1.08s, and 1.39s for sliding window sizes 5, 10, 15, and 20, while ReLIM required 0.43s, 0.73s, 1.02s, and 1.31s on Webdocs dataset. Similar results are also obtained for the Accidents dataset. From these results we conclude that FPCFI require significantly higher time usage compared to ReLIM.