Financial time series pattern matching with extended UCR Suite and Support Vector Machine

Abstract

Chart patterns are frequently used by financial analysts for predicting price trends in stock markets. Identifying chart patterns from historical price data can be regarded as a subsequence pattern-matching problem in financial time series data mining. A two-phase method is commonly used for subsequence pattern-matching, which includes segmentation of the time series and similarity calculation between subsequences and the template patterns. In this paper, we propose a novel approach for locating chart patterns in financial time series. In this approach, we extend the subsequence search algorithm UCR Suite with a Support Vector Machine (SVM) to train a classifier for chart pattern-matching. The experimental results show that our approach has achieved significant improvement over other methods in terms of speed and accuracy.