A comparative study of the various optimizations applied on an LSTM for the application of Stock Price Predictions

Predicting time series data can prove to be challenging. However, recent developments in human technology have provided us with certain tools to help us with this particular task reliably and without much fear of loss. The long short term memory (LSTM) model provides us with a pathway to better predictions and hence better results. The results can further be improved by using various optimization techniques. This paper presents the usage of an LSTM model with the different kinds of available optimization techniques on stock datasets retrieved from NIFTY 50 and Yahoo Finance. The results are compared with respect to the performance of each optimization technique. The mean absolute error (MAE), and root mean square error (RMSE) are the metrics used to evaluate which LSTM model paired with which optimization technique is best. We find that the error is found to be least in the Adam, and Adamax optimizations deeming them the most suitable for this application.