A High frequency stock price prediction model based on Boosting and information entropy weighted LSTM neural network

Abstract

With the continuous development of data acquisition technology, high-frequency time series forecasting, such as stock prices, has become a hot issue. This paper proposes a LSTM neural network based on Boosting and information entropy weighting. The proposed method extracts the functional features of high-frequency time series by using orthogonal polynomial expansion, and uses the Boosting frame to recursively fit the residual predicted by LSTM neural network. Considering that the dimension of the prediction variable of LSTM neural network is a super parameter, we propose a model average method based on information entropy weighting, which theoretically balances the variance and bias of the prediction model. In addition, the basic model in the Boosting framework can be arbitrary, which greatly expands the applicability of the proposed method. The results of real data analyses show that the proposed method can effectively improve the prediction accuracy of the original LSTM neural network and has robustness. Finally, the proposed method can be further applied to the real-time monitoring of trace elements, the monitoring of road traffic flow and the prediction of daily average temperature curve in environmental science.