Feature-Ranked Attention-Driven LSTM: A Novel Hybrid Model for Short-Term Stock Price Predictions

Accurate financial market price forecasting helps investors make financial decisions and governments formulate macro-financial policies. To accurately predict the price trend of stocks, a hybrid model combining Random Forest (RF) and Long Short-Term Memory (LSTM) network based on Attention Mechanism (AM), called RF-AM-LSTM, is proposed. Firstly, the input features, such as raw historical data, technical indicators and broad market indices, are ranked in importance and the optimal feature subset is filtered by RF. Then, the filtered feature sequences are processed by an AM to increase attention to important features and decrease attention to redundant features. Finally, the processed sequences are input to the LSTM network for the closing price prediction task. Three stocks listed in the Chinese A-share market are selected for stock price forecasting to evaluate the performance of the hybrid model. The experimental results show that the RF-AM-LSTM can effectively achieve short-term stock price forecasts and achieve the best prediction results on all four regression evaluation indicators.