Time Series Clustering Based on ICA for Stock Data Analysis

Time series clustering is an important task in time series data mining. Compared to traditional clustering problems, time series clustering poses additional difficulties. The unique structure of time series makes many traditional clustering methods unable to apply directly. This paper presents a novel feature-based approach to time series clustering, which first converts the raw time series data into feature vectors of lower dimension by using ICA algorithm, and then applies a modified k-means algorithm to the extracted feature vectors. Finally, to validate effectiveness and feasibility of the presented method, we use it to analyze the real world stock time series data and achieve reasonable results.