Time Series Forecasting in Financial Market: Long Short-Term Memory (LSTM) Approach for Stock Price Prediction

Stock market nature is nonlinear, and research on it has become increasingly important in recent years. The influence of numerous factors on stock prices renders stock forecasting an intricate and demanding task. Possible stock market price forecast increases people's gains while minimizing their risks. Ma-chine learning and deep learning techniques play an important role in forecasting stock prices in today's scenario, considering technical indicators. The choice of deep learning technique and tweaking its hyper-parameters are crucial components of stock price prediction. Several model architectures are suggested here, taking into account LSTM deep learning techniques and hyper-parameters like the batch size, number of hidden layers and number of epochs. Hyper-parameters such as the number of years data, the number of preceding days considered, and database spitting are examined and discussed here, in addition to hyper-parameters relating to LSTM networks. The model with 11 years of historical data, 60 or 100 previous days consideration, and 70:10:20 database split outperforms as compared to other model architectures suggested in the paper.