China’s commercial bank stock price prediction using a novel K-means-LSTM hybrid approach

Abstract

China’s commercial Bank shares have become the backbone of the capital market. The prediction of a bank's stock price has been a hot topic in the investment field. However, the stock price is always unstable and non-linear, challenging the traditional statistical models. Inspired by this problem, a novel hybrid deep learning approach is proposed to improve prediction performance. By modifying the distance measurement algorithm into DTW, an improved K-means clustering algorithm is proposed to cluster out banks with similar price trends. Then those clustered stocks are used to train a long and short-term memory (LSTM) neural network model for static and dynamic stock price prediction. Besides, by transforming the output of the LSTM network into multi-step output to predict multi-time intervals at one time, the performance of the long-term forecasts is improved. Through experiments, it is found that the hybrid model performs better than the single model in generalization ability and accuracy(i.e. R-SQUARE, MAE, MSE). Moreover, the multi-step output static prediction outperforms the dynamic rolling prediction for long-term prediction. In summary, this approach can predict stock prices more accurately and help investors and companies to make more profitable decisions.