Forecasting time series data using moving-window swarm intelligence-optimised machine learning regression

摘要

This study proposes a hybrid time series forecast model namely a moving-window firefly algorithm (FA)-based least squares support vector regression (MFA-LSSVR). In the proposed model, the LSSVR captures patterns of historical data and predicts future values of time series data while the FA is used to optimise the LSSVR`s parameters to improve the predictive accuracy. The proposed model was trained and tested using two actual datasets of the daily energy demand data and the stock price data. Experimental results show that the proposed MFA-LSSVR model is effective in forecasting time series data and the comparison results revealed that the proposed model outperforms other models, i.e., the LSSVR and the ARIMA (autoregressive integrated moving average) in predicting energy demand and stock price. This study's findings, thus, provide decision makers a potential approach in early forecasting future patterns of time series data.