Similarity Pattern Discovery Using Calendar Concept Hierarchy in Time Series Data

Abstract

Most existing approaches for similarity search did not consider applying calendar concept hierarchy to search for similar patterns from time series data. In this paper, we present two techniques that capture scale oriented features of time series and provide an analyzing method for the multi-resolution view along the time dimension. Especially, we propose that a similarity search which makes the most of calendar concept hierarchy involves three stages which consist of data cube count based on time concept hierarchy, sequence division by time level and feature vector extraction. Because these feature vectors are inserted into multi-dimensional index, pre-processing step executes only one time at the beginning of the search process without adding considerable computing cost. Finally, we show that the proposed techniques find useful knowledge with low computational complexity and discovered rules can be applied to industrial fields.