Comparative Evaluation of Standard, Bayesian and Quantile LSTM Networks for Time-Series Forecasting of Indian Stock Prices

Long Short-Term Memory (LSTM) neural networks have recently gained prominence as one of the most accurateand reliable deep learning architectures for time-series forecasting. Despite their growing popularity, there remainslimited comparative research on different variants of LSTM models within financial forecasting contexts. This studyaims at bridging this gap through a comprehensive evaluation of three LSTM-based approaches—Standard LSTM,Bayesian LSTM, and Quantile Regression LSTM—applied to the prediction of NIFTY-10 stock prices in the Indianstock market. Through a detailed comparison of predictive accuracy, uncertainty quantification, and directionalperformance, this work identifies the most robust and consistent LSTM model for stock market forecasting.