Research On Stock Price Trend Clustering Based On Time Series Decomposition Model

The k-shape clustering algorithm is widely used in the field of medical data,such as electrocardiogram analysis.Different from the DTW algorithm,this algorithm uses SBD(Shape Based Distance)distance as its distance measurement method,so that it is closer to the extraction of shape factors and can often discover the local similarity of sequences.Existing methods for trend prediction and analysis of financial time series include BP neural network,LSTM and so on.The main research content of this article is to apply k-shape clustering algorithm to stock price trend prediction.The STL time series decomposition algorithm is used to extract its long-term trend terms and denoise.The author believes that the sequences of the same cluster can be used to compare and even predict the similarity of the trend.In empirical analysis,using cluster classification labels as one of the input layer parameters of BP neural network can improve its prediction accuracy.In order to prove the reliability of the results,multiple sets of controlled experiments were set up in this paper,including the development of a hierarchical clustering algorithm based on the SBD distance,and the cluster classification labels of the hierarchical clustering results were also used as one of the parameters for neural network training.In order to make the k-shape algorithm achieve better performance in highly uncertain financial time series,this paper improves its cluster center extraction algorithm and normalization steps.The experimental results show that the improved k-shape algorithm can better cluster the stock price trend data,and the cluster centers of each cluster are also closer to the shape center.