Stock Price Forecast Based on Dueling Deep Recurrent Q-network

Stock prices belong to time series data, and accurate prediction of stock prices is a difficult problem. The traditional methods of predicting stocks through statistics or regression models usually do not consider the interrelationship and complex relationship between various variables in the stock market, which may lead to information loss and error. To overcome these challenges, this paper studies a deep reinforcement learning algorithm based on Dueling Deep Recurrent Q-Network (Dueling DRQN), which improves the accuracy of stock price prediction by combining the advantages of RNN models in processing sequence data with the Q algorithm's ability to adaptively learn and predict stock price patterns and trends. Specifically, we first embed Long Short-Term Memory (LSTM) layers into the DQN network to model sequence data, and then use the Dueling architecture to make the model more generalizable and interpretable. The experiment uses a variety of stock information to predict the performance of stocks, taking the total return of the test set as the only evaluation index, and the specific results show that the trained Dueling DRQN model performs well in stock price prediction, and the Dueling DRQN performs better than other models. The total return under different test datasets can exceed 90%.