Blended computation of machine learning with the recurrent neural network for intra-day stock market movement prediction using a multi-level classifier

Abstract

Intra-day stock market forecasting is one of the challenging tasks as its nature is difficult to predict due to the non-volatility of the stock market. Therefore, the building block concepts of machine learning techniques with the recurrent neural network have been used. These techniques are very supportive for developing the long short-term memory (LSTM) network. This network is mainly used for sequential learning while performing classification. This paper presents research on applied computing to assess the efficacy of a multi-level classifier over single classifier. The multi-level classifier is hybrid of machine learning techniques (logistic regression classifier, decision tree, support vector machine) and recurrent neural network (LSTM) aiming to maximize the predictive ability, increases the interpretability, minimizes the complexity, reduced the training time, and provides the generalized platform with an optimistic forecasting model. The developed model has the potential to act as an intelligent system with the ability to adapt the market behavior present in the dataset which analyzes the best performing feature subset based on model accuracy. The comparative results clearly reveal that the blended computation of multi-level classifiers outperforms the existing works and it also leads to more precise model development with the maximum predictive ability, approximately 10–12% increment in accuracy.