Stocks and models: A comparative benchmark of machine learning and deep learning approaches for stock market prediction

The stock market, a cornerstone of global finance, plays a critical role in connecting investors with companies seeking capital through Initial Public Offerings (IPOs). Despite its opportunities for wealth generation, the market’s inherent volatility and com- plexity pose significant challenges for accurate prediction. To address these limitations, this study explores the growing impact of machine learning (ML) and deep learning (DL) in stock market forecasting. The findings aim to inform investors, researchers, and practitioners in selecting effective tools for financial forecasting amidst the uncertainties of the stock market. This paper benchmark the performance of three prominent time-series models such as Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and eXtreme Gradient Boosting (XGBoost) by using 76203 dataset of historical OHLCV (Open, High, Low, Close, Volume) stock data. This study finds that while XGBoost offers the fastest training and inference times, LSTM consistently outperforms both GRU and XGBoost in terms of forecasting accuracy compared to GRU and XGBoost. Specifically, XGBoost achieves lower error values across key metrics such as RSME, MAE, MAPE, and higher R² scores, though the differences among the models remain relatively close.