Stock price prediction based on LSTM and LightGBM hybrid model

Finding an accurate, stable and effective model to predict the rise and fall of stocks has become a task increasingly favored by scholars. This paper proposes a long short-term memory (LSTM) network based on Pearson's correlation coefficient and a Bayesian-optimized LightGBM hybrid model, named as LSTM-BO-LightGBM, to solve the problem of stock price fluctuation prediction. The multilayer bidirectional LSTM-BO-LightGBM prediction model is compared with the LSTM-BO-XGBoost hybrid model, the LSTM-LightGBM hybrid model, the LSTM-XGBoost hybrid model, the single LSTM network model and the RNN network model. The prediction result of the LSTM-BO-LightGBM model for the "ES = F" stock is an RMSE value of 596.04, MAE value of 15.24, accuracy value of 0.639 and f1_score value of 0.799, which are improved compared with the prediction results of the other five models. At the same time, when applying the model to "YM = F", "CL = F", "<^>TNX", "<^>N225", "NQ = F", "AAPL", "GC = F", "JPY = X" and "SI = F", all of the nine stocks showed good forecasting performance. The results demonstrate that the multilayer bidirectional LSTM-BO-LightGBM model proposed in this paper has better approximation ability and generalization ability in the stock fluctuation forecast than previous models and can well fit the stock fluctuation.