A Novel Similarity Measure Framework on Financial Data Mining

Similarity measure is one of the core problems in data mining research on time series. Both Euclidean Distance and Dynamic Time Warping methods are widely used in similarity measure, which have their own limitations. Based on the symbolic time series data, this paper creatively brought forward a new kind of similarity measure framework SFVM (SAX Feature Vector Model). Based on the architecture of SFVM, we introduced three feature extraction methods, which can be extended as three new similarity measure methods. The paper obtained stock data from the Standard & Poor´s 500 index, and compared the behavior of several similarity measure methods in clustering research. The experiment results showed that the three methods could efficiently measure the similarity according to the whole trend by comparing with the classic methods.