A Novel Weighted Frequent Pattern-Based Outlier Detection Method Applied to Data Stream

摘要

Against the problems that the weight value will seriously influence the result of outlier detection, this paper proposes a weighted frequent pattern-based outlier detection method called WFP-Outlier to discover the implicit outliers from the weighted data stream, where the WFP-Outlier method is conducted in two phases. In the weighted frequent pattern mining phase, we use the "pattern extension" operation to extend the "weighted frequent" k-patterns to (k+1)-patterns, and the maximal weight value is used in the mining process to improve the mining efficiency. In the outlier detection phase, based on the mined weighted frequent patterns, we defined two abnormal indexes named Weighted Frequent Pattern Abnormal Index (WFPAI) and Transaction Abnormal Index (TAI) to measure the abnormal degree of the detected transactions, and the transactions with small TAI values are judged as the abnormal transactions. The experimental results on a synthetic dataset show that the proposed WFP-Outlier method is suitable detecting the implicit outliers for the weighted data stream, and the detection efficiency is higher than that of FindFPOF method, OODFP method and LFP method.