Integrating Harmony Search Algorithm and Deep Belief Network for Stock Price Prediction Model

摘要

Stock price predicting is of great practical value in stock market trading and is the foundation of programmatic trading and high-frequency trading. Deep Belief Network (DBN) has been widely used in the field of stock market predicting. However, due to the lack of effective parallel training, empirical methods are still used in the application to determine the important network structure parameters in DBN, resulting in high calculation cost and low efficiency. This study puts forward a hybrid model of integrating metaheuristic Harmony Search (HS) algorithm and DBN. By setting up the self-learning mechanism of the network, the feedback evaluation of objective function is applied to let the network itself calculate the most appropriate structural parameters and improve the model prediction performance. Through the experimental verification on the 160 historical trading days of the Shanghai Composite Index, the mean absolute percentage error (MAPE) and return profit margin of the hybrid model presented in this paper are 1.56% and 16.04%, and the overall performance is better than the existing stock price prediction models.