LSTM Based Recurrent Enhancement of DQN for Stock Trading

In real stock trading tasks, technical and fundamental analysis are required in order to ensure positive earning at the end of trade. Machine learning approaches are widely used in financial market analysis especially stock price prediction to support the investor decision. However, most of the techniques only provide price forecasting without action recommendation. Reinforcement learning technique facilitates stock trading decision using algorithm that can make decision to maximize profit. In this work, a novel approach based on LSTM as a preprocessing recurrent network to enhance the Deep Q-Network (DQN) is presented. The experiment results promote the robustness of the proposed algorithms and it can be considered as a baseline algorithm for other agent-based stock trading research.