The Method of Analysis Granularity Determination for Multi-granularity Time Series

Abstract

Select a proper analysis granularity for time series is fundamental for data mining and knowledge discovery. In this paper, we propose the method of analysis granularity determination for multi-granularity time series. We first give the strategy of candidate analysis granularity set. Then through sampling the original time series at the granularity of the set, we calculate the missing rate and information integrity degree to evaluate the quality of the candidate analysis granularity. Finally, experiments have been run on medical dataset and the experimental results show that the proposed algorithm can achieve better performance for multi-granularity time series.