Periodicity-Oriented Data Analytics on Time-Series Data for Intelligence System

Abstract

Periodic pattern mining models analyze patterns which occur periodically in a time-series database, such as sensor readings of smartphones and/or Internet of Things devices. The extracted patterns can be utilized for risk prediction, system management, and decision-making. In this article, we propose an efficient periodicity-oriented data analytics approach. It ignores intermediate events deliberately by adopting the concept of flexible periodic patterns, so it can be applied to more diverse real-life scenarios and systems. Moreover, the proposed approach adopts a novel symbol-centered data structure instead of existing data structures for state-of-the-art approaches of periodic pattern mining. Performance evaluations on real-life datasets, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Diabetes, Oil Prices</i> , and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Bike Sharing</i> , and requirements show that our approach has better runtime, memory usage, number of visited patterns, and sensitivity than efficient periodic pattern mining (EPPM) and flexible periodic pattern mining (FPPM), which are the state-of-the-art approaches in the same field. The experimental results show that the proposed algorithm will require less runtime and smaller memory than the existing algorithms on most data and requirements in real life.