Stock Price Forecasting Using Hybrid Prophet—LSTM Model Optimized by BPNN

Abstract

AbstractStock price forecasting is one of the important fields of research owing to the commercial applications in the present scenario. The investment in the stocks is high these days due to the lucrative returns. Even though the risk element is high, people prefer investing in high stakes. As the investment is done on a higher scale, it is obvious that the investor’s anxiety too increases. And due to this reason, the price movement forecasting in the stock market has been and is still a major challenge to be faced. Various models have been proposed in the literature for prediction of stock prices, but it suffers from various issues such as accurate prediction, high complexity, and low speed convergence. Thus, in this work, a hybrid model which is a combination of prophet and long short-term memory (LSTM) is proposed to overcome these issues. Again at the end, forecasting is further optimized by backpropagation neural network (BPNN). The prophet model uses linear and nonlinear data to predict the stock prices, but still some residuals remain corresponding to the nonlinear data. Here, this nonlinearity in the data is mitigated by the LSTM model. The performance parameters used are root mean square error (RMSE), mean absolute percentage error (MAPE), and mean average error (MAE). This model will be very much suitable for forecasting the stock prices of various stock exchanges for all over the globe.KeywordsForecastingProphet modelLSTM modelHybrid modelRoot mean square error