Research on Maximal Frequent Pattern Outlier Factor for Online High-Dimensional Time-Series Outlier Detection

Frequent pattern outlier factor is used to detect outliers with complete frequent itemsets. But it is difficult in real-world time-series data streams application because of its low efficiency. In this paper, we propose a novel maximal frequent pattern outlier factor (MFPOF) and an outlier detection algorithm (OODFP) for online high-dimensional time-series outlier detection. Firstly, the time-series data streams are processed with sliding window to discover maximal frequent itemsets. Then the frequent patterns are simplified to compute the MFPOF of time-series data streams. Experimental results show that our approach not only provides higher efficiency, but also equivalent accuracy.