Machine Learning in Financial Markets: A Performance Comparison of Basic and Stacked LSTM Models

Stock price prediction plays a crucial role in financial markets, enabling investors and analysts to make data-driven decisions. The effectiveness of Long Short-Term Memory (LSTM) models, specifically the Basic LSTM Model and the Stacked LSTM Model, is evaluated for forecasting stock prices using 10 years of daily Nifty 50 index data (2015–2024). A time-series analysis approach is applied, where the dataset undergoes preprocessing using MinMax Scaling, and predictions are generated for 30-day, 60-day, and 90-day forecasts. The models are assessed based on key performance metrics, including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and R-squared (R²) values, to determine prediction accuracy. The findings indicate that the Stacked LSTM Model consistently outperforms the Basic LSTM Model, particularly in 30-day and 90-day forecasts, by demonstrating higher accuracy and lower error rates. In contrast, the Basic LSTM Model exhibits higher deviations in 60-day forecasts, making it less reliable for medium-term predictions. Despite the effectiveness of LSTM-based models in capturing stock price trends, challenges such as market volatility, computational complexity, and hyperparameter sensitivity remain. Future advancements, including hybrid deep learning models, sentiment analysis, and real-time forecasting, could enhance predictive accuracy, making LSTM-based models a valuable tool for financial forecasting.