Combine holts winter and support vector machines in forecasting time series

Abstract

This study proposes on a combine methodology that exploits the Holts- Winter (HW) model and the Support Vector Machines (SVM) model in forecasting time series. Problems of forecasting using time series data have been and still being addressed at every sphere of research using different approaches. The performance of the forecast was compared among the three models, the HW model, the SVM model and the combine model (HW and SVM). Four different data sets namely, airline passengers’ data, machinery industry production data, clothing industry data and sugar production data were considered in the study. The statistical measures such as mean squared error (MSE), mean average error (MAE) and correlation coefficient, R, were used to evaluate the performance of the propose model. The result of this study indicated that the combine model shows an improvement of 149.3% over HW model and 35.9% improvement over the SVM model for the airline passengers’ data. The result of the machinery industry presented that the combine model shows an improvement of 93.3% over HW model and 42.8% improvement over the SVM model. In the case of the clothing industry the result shows the combine model gives an improvement of 61.6% over HW model and 12.0% improvement over SVM model. Lastly, with respect to the sugar production, the result shows that the combine model indicated an improvement of 34.4% over HW model and 25.1% improvement over SVM model. Therefore the results of the experiments suggest that the proposed combine model is more reliable in time series when compared with the individual models.