Stock market prediction using optimized long-short term memory based on improved salp swarm optimization

The prediction of stock price volatility is thought to be one of the most fascinating and important study topics in the financial sector.  Deep learning (DL) uses cutting-edge computing technology to analyze data, particularly in the discipline of finance, to deliver insightful analysis.  Due to the advantages of sequential learning, the LSTM network has demonstrated notable performance when it comes to time series prediction.  It might be very difficult for a researcher to build and choose the best computationally optimal LSTM network for stock market forecasting.  The model's ability to learn is impacted by multiple hyperparameters that it must control due to the nature of the data. In addition, several earlier research used heuristics based on trial and error to guess these parameters, which may not always result in the best network.  Furthermore, the hyper-parameter values have a significant impact on LSTM model accuracy.  To increase the efficiency and precision of the stock prediction method, an improved salp swarm optimization (ISSA) algorithm is used in this research to find satisfactory LSTM model parameters.  The ISSA approach uses partial opposition-based learning (POBL) to enhance the population diversity and avoid local optima problems.  To verify the ability of the proposed ISSA-LSTM prediction method, four different stock market datasets are considered for experiments.  The experimental results confirmed that the developed optimized ISSA-LSTM approaches produced high prediction accuracy and fast convergence rate.