MBPD: Motif-Based Period Detection

Abstract

AbstractMassive amounts of data are generated daily at a rapid rate. As a result, the world is faced with unprecedented challenges and opportunities on managing the ever-growing data. These challenges are prevalent in time series for obvious reasons. Clearly, there is an urgent need for efficient solutions to mine large-scale time series databases. One of such data mining tasks is periodicity mining. Efficient and effective periodicity mining techniques in big data would be useful in cases such as finding animal migration patterns, analysis of stock market data for periodicity, and outlier detection in electrocardiogram (ECG), analyses of periodic disease outbreak etc. This work utilizes the notion of time series motifs for approximate period detection. Specifically, we present a novel and simple method to detect periods on time series data based on recurrent patterns. Our approach is effective, noise-resilient, and efficient. Experimental results show that our approach is superior compared to a popularly used period detection technique with respect to accuracy while requiring much less time and space.KeywordsMotifPeriodTime seriesBig data