Authors: Shaikh Sahil Ahmed; Mahesh Kankar; Nagaraj Naik; Biju R. Mohan
Addresses: National Institute of Technology, Karnataka, India ' National Institute of Technology, Karnataka, India ' National Institute of Technology, Karnataka, India ' National Institute of Technology, Karnataka, India
Abstract: Forecast analysis is in very high demand in many fields for improving sales and operation planning in various industries and enterprises. So, accuracy is a significant factor in forecasting stock market prices. We already know there are existing deep learning models for stock market prediction such as gated recurrent unit (GRU), support vector machine (SVM), multilayer perceptron (MLP), etc. This paper enhanced the prediction of stock prices using series hybrid models over single deep learning models. The models we used are autoregressive integrated moving average (ARIMA), deep belief network (DBN), long short-term memory (LSTM), and performed analysis on hybrid models in comparison with single models. We have chosen a model as ARIMA, LSTM, and hybrid as ARIMA-DBN and ARIMA-LSTM. For finding the best fit parameter for ARIMA and DBN, the particle swarm optimisation (PSO) technique is used. We compared the various models based on performance errors like MSE, RMSE, MAPE, etc. As already existing ARIMA and LSTM is not good enough for forecasting and so we worked over the ARIMA-DBN model to overcome the limitations of other models. After research, we found out that series hybrid ARIMA-DBN is effectively better than other single models for stock market prediction.
Keywords: deep learning; time series forecasting; autoregressive integrated moving average; ARIMA; linear and nonlinear models; particle swarm optimisation; PSO.
International Journal of Swarm Intelligence, 2022 Vol.7 No.1, pp.66 - 81
Received: 29 Jun 2020
Accepted: 16 Dec 2020
Published online: 24 Feb 2022 *