Short-term Forecast of Stock Price of Multi-branch LSTM Based on K-means

摘要

In order to improve the short-term forecast accuracy of the stock, this paper first uses the time window to divide the stock price sequence data in a period of time into a number of equal sub-sequences. The K-means algorithm is used to cluster the stock price sub-sequences .Then according to the number of clusters, The same number of long short-term memory(LSTM) neural network models are constructed with the clustering results used to train the corresponding LSTM model and a multi-branch LSTM stock price short-term forecasting model is constructed. When the stock price of one day is predicted, the distances between the stock price sequence of the neighboring days and the centers of the K-means cluster are calculated, and then the stock price sequence of the neighboring days is inputted into the LSTM forecasting model corresponding to the shortest distance clustering center to predict the stock price. The experimental results show that the proposed model has higher prediction accuracy than the BP neural network prediction model and the single LSTM neural network prediction model, which is based on the actual closing price of one stock.