Enhancing Stock Price Predictions Through LSTM-based Recurrent Neural Networks

Analyzing finances has become increasingly challenging in today's investment landscape, where making valuable and informed investment decisions is crucial. The fluctuation of share prices plays a pivotal role in determining investors' profits or losses. Current forecasting techniques encompass both linear and non-linear algorithms. However, these methods primarily emphasize forecasting changes in the stock index or forecasting prices for individual companies based on their daily concluding rates. The proposed methodology introduces a model-agnostic method. Rather than conforming data to a specific model, this approach aims to identify unseen trends inherent in the data through deep learning models. In this study, three distinct deep learning models—HMM, RNN, and LSTM—are employed to predict prices using the dataset from ICICI Bank. Their performances are compared, revealing that the LSTM model adeptly discerns evolving trends. The LSTM model displays the lowest error percentage at 2.36%, outperforming other models such as HMM (7.32%) and RNN (3.94%).