Date of Award:

8-2021

Document Type:

Thesis

Degree Name:

Master of Science (MS)

Department:

Computer Science

Committee Chair(s)

Nicholas S. Flann

Committee

Nicholas S. Flann

Committee

Vladimir Kulyukin

Committee

Chad Mano

Abstract

In this study, we are developing a reinforcement learning-based strategy for the intraday trading of stocks. This study’s primary goals include developing an environment that can be used as a simulator for the day trading stock market and, train an agent to trade in this environment by performing actions and finding the optimal policy to maximize its reward.

This study also focuses on experimentation with different state representations to understand how data representation affects the learning system. We have experimented with three different state representations. All the representations focus on presenting the intraday stock data as images. We have chosen images rather than numerical data as our state representative to make it human interpretable. We want the agent to learn the insights from patterns and trends and apply that knowledge to make a profit, just like a seasoned human trader. We have also experimented with two different learning algorithms, the effects of exploration rate decay and the effects of reward penalty.

Checksum

f6884d9d6b2d88027f0165c0a21ee6ed

Available for download on Saturday, August 01, 2026

Share

COinS