Decision support framework for the stock market using deep reinforcement learning

Abstract

In stock markets, investors adopt different strategies to identify a sequence of profitable investment decisions to maximize their profits. To support the decision of investors, machine learning (ML) software is being applied. In particular, deep learning (DL) approaches are attractive since the stock market parameter presents a highly non-linear behavior, and since DL techniques can track short time and long time variations. In contrast to supervised ML techniques, deep reinforcement learning (DRL) gathers DL’s benefits and adds the real-time adaptation and improvement of the machine learning model. In this paper, we propose a decision support framework for the stock market based on DRL. By learning the trading rules, our framework recognizes patterns, maximizes the profit obtained, and provides recommendations to the investors. The proposed DRL framework outperforms the state-of-the-art framework with 0.86 % of F1 score for buy operations and 0.88 % of F1 score for sale operations in terms of evaluating the positioning strategy.