Novel Deep Learning Hybrid-Ensemble Method for Financial Domain Information and Stock Price Forecasting

This study aims to develop the novel technique for time series forecasting of stock prices. For the research methods, we developed method based on stacking mechanism of LSTM(Long Short Term Memory), Ridge Regression, Linear Regression and CNN(Convolutional Neural Network) for predicting “Close” prices of Amazon, Microsoft, Netflix and Apple stock prices for next day, next 7 days, next 14 days and 21 days. In the results of this research, we compared our method to another popular hybrid method CNN-LSTM and also to base models of our method such as CNN and LSTM as well as RNN(Recurrent Neural Network). We used four datasets and four periods to predict and our method outperformed other models on metrics such MSE (Mean Squared Error), RMSE (Root Mean Squared Error), RMSE (Root Mean Squared Error) and SD (Standard Deviation of Errors). In conclusion, the LSTM-ENSEMBLE-CNN (proposed method) model stands out as the best performer among the tested models. It achieves the lowest MSE, RMSE, and MAE, indicating minimal prediction error.