Deep Learning Techniques GAN and LSTM for Stock Market Closing Price Prediction

Abstract

Objective: The goal of this study is to enhance   forecast   accuracy   by   leveraging   advanced deep learning (DL) methods, specifically Generative Adversarial Networks (GAN) and Long Short-Term Memory (LSTM) networks.

Methods: Using   historical stock data and 32 influencing features, GANs   and   LSTMs   are   trained   to   predict   the   stock   price   using   the   RMSE, with 80% of the data being training and 20% testing.

Results: The GAN model achieved the lowest RMSE of 5.36 on the testing dataset, outperforming both LSTM and traditional ARIMA models. The LSTM model also showed robust performance with an RMSE of 6.6. These results indicate the effectiveness of GANs in capturing complex patterns in stock price data.

Conclusion: Despite computational challenges, the GAN model's superior accuracy underscores its potential for stock price prediction. This study highlights the value of integrating diverse features and DL models for financial forecasting, suggesting future research directions including sentiment analysis and hybrid model development for improved predictive performance.