LSTM Neural Network Model with Feature selection for Financial Time series Prediction

Abstract

The case of features selection plays an important role in fine-tuning the prediction capacity of machine learning models. This paper reviews the different scenarios with three sets of features in each case and evaluate the training and validation data performance with and without these features. How the prediction results change can be seen as and when the different features are included or excluded and Recursive feature elimination, Correlation, Random forest algorithm is used for feature importance and evaluate the results with LSTM networks.