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

Abstract

Accurate stock price prediction is vital for investors and analysts to make informed decisions and mitigate risks. Traditional methods like ARIMA struggle with the complex, non-linear patterns of stock market data, leading to less precise forecasts. This study addresses the gap by leveraging advanced deep learning methods, specifically Generative Adversarial Networks (GANs) and Long Short-Term Memory (LSTM) networks, for stock price forecasting. Using historical stock data and 32 influencing features, we trained GANs and LSTMs, with Root Mean Square Error (RMSE) as the evaluation metric. The data was split into 80% training and 20% testing. Our results show that the GAN model achieved a lower RMSE of 5.36, outperforming the LSTM model (RMSE of 6.6) and traditional ARIMA models, demonstrating GAN’s superior ability to capture complex patterns. While computational challenges exist, GAN’s accuracy underscores its potential for stock price prediction. Future research could explore hybrid models and sentiment analysis to further improve prediction accuracy.