Features extraction and correlation analysis of stock index

Time series exists in lots of fields, therefore data mining in time series has important research value. Considering correlation analysis is the foundation of time series data mining, the paper concentrates on the topic. We choose robust Dynamic Time Warping (DTW) distance and propose the improvement for standard DTW algorithm to deal with its large computing time cost: extracting the feature points according to fluctuation at first and organizing the features in a binary tree. It reduces the dimension and meanwhile reserves trend information. DTW with a computing window is then employed on the feature sequence. Experiments on three datasets and two scenarios in Shanghai Stock Market closed price series show that, the new method is much faster with keeping high accuracy as well.