Stock market Prediction using Reinforcement Learning Technique

Abstract

Several recent studies have attempted to construct efficient mechanical trading systems utilizing machine learning for stock price prediction and portfolio management. However, these systems have a drawback in that they are primarily focused on supervised learning, which is insufficient for situations involving long-term goals and delayed rewards.Financial markets, which include the stock market, are vital for the growth of capitalism since they are marketplaces where trading takes place. The stock market is essentially the ownership of a portion of publicly traded corporations in the form of shares, which may be sold to claim ownership. To accurately estimate a company's stock price, one must devote time to it in order to make the optimal trade. Our results show that the Double Q-Learning Network performs better than the Deep Q-Learning Network and Dueling Double Q-Learning Network by receiving the highest reward in each state where the agent tries to make the best decision, i.e., buying when the market is moving upwards, selling when the market is moving downwards, and holding back when the market is moving unstable. Keywords - Stock price prediction, Reinforcement learning, Q learning, Deep Q-learning Network, Double Q-Learning Network, Dueling Q-Learning.