A Qualitative Feature Extraction Method for Time Series Analysis

Time series feature extraction is a way to reveal the most important characteristics of a (or a set of) time series. It is an effective pre-processing step for many time series mining tasks such as clustering and indexing. In this paper, we propose a new qualitative feature extraction method. The method differs from most available methods in that it mainly focuses on the shape, instead of the actual values, of any time series. In the proposed method, a set of shape oriented patterns is defined and the feature of a data sequence is referred to as the combination of these patterns. A procedure for identifying patterns in a given sequence is developed. Experiments on real stock price data are performed to evaluate the performance of the proposed method used for clustering and similarity search.

1 INTRODUCTION

In recent years, there has been a lot of research interest in time series data mining [1]–[16]. A time series is a sequence of values which could represent digital signals, stock prices, temperature records, or medical measurements. One of the most important tasks in time series mining is to find sequences from a large database which are similar in a certain sense to the given query sequence.