Comparison of hybrid and non-hybrid models in short-term predictions on time series in the R development environment

Abstract

Because many time series usually contain both linear and nonlinear components, a single linear or nonlinear model may be insufficient for modeling and predicting time series. Therefore, estimation results are tried to be improved by using collaborative models in time series short-term prediction processes. In this study, the performances of both stand-alone models and models whose different combinations can be used in a hybrid environment are compared. The mean absolute percentage error (MAPE) metric values obtained from two different categories were evaluated. In addition, the estimation performances of three different approaches such as equi-weighted (EW), variable-weighted (VW) and cross-validation-weighted (CVW) for hybrid operation were also compared. The findings on the container throughput forecast of the Airpassengers dataset reveal that the hybrid model's forecasts outperform the non-combined model.Because many time series usually contain both linear and nonlinear components, a single linear or nonlinear model may be insufficient for modeling and predicting time series. Therefore, estimation results are tried to be improved by using collaborative models in time series short-term prediction processes. In this study, the performances of both stand-alone models and models whose different combinations can be used in a hybrid environment are compared. The mean absolute percentage error (MAPE) metric values obtained from two different categories were evaluated. In addition, the estimation performances of three different approaches such as equi-weighted (EW), variable-weighted (VW) and cross-validation-weighted (CVW) for hybrid operation were also compared. The findings on the container throughput forecast of the Airpassengers dataset reveal that the hybrid model's forecasts outperform the non-combined model.