A comprehensive review on multiple hybrid deep learning approaches for stock prediction

Numerous recent studies have attempted to create efficient mechanical trading systems through the use of machine learning approaches for stock price estimation and portfolio management. Using the ability to foresee the future trends of the stock performance, the return of investment can be maximized for short-term trading. This paper will review various Artificial Intelligence (AI) and Machine Learning (ML) strategies for stock price forecasting. The aim of this review is to discuss various techniques for stock price prediction that incorporate ARIMA, LSTM, Hybrid LSTM, CNN, and Hybrid CNN. Additionally, it will also discuss the limitations and accuracy of the various models, including the ARIMA model, the LSTM model, the MI-LSTM model, the Bi-LSTM model, the LSTM-DRNN model, the CNN model, the GC-CNN model, the CNN-LSTM model, the CNN-TLSTM model, and the CNN-BiLSTM model, in terms of percentage of accuracy or error calculation in terms of standard accuracy measures like RMSE, MAPE, MAE. The models can be used to forecast either the accurate stock rate, induced by the low MSE, RMSE and MAE of LSTM models, or the general trend and deflection range of the stock the following day, induced by the ability to dynamically capture swift changes in the system of CNN models. These characteristics consequently illustrate the advantages of the hybrid model at efficiently and accurately forecasting stock attributes.