A Weight-adjusting Approach on an Ensemble of Classifiers for Time Series Forecasting

In this study, we present a hybrid heterogeneous forecasting model that combines autoregressive integrated moving average (ARIMA) model and two machine learning classifiers (i.e. support vector machine (SVM) and artificial neural network (ANN)) for time series forecasting. The approach adjusts each model's weight based on their ability and history of predicting numerical values. A weighted numerical value based on each model's numerical output and their weight is calculated as the final output. An air quality dataset is used to evaluate our approach. We conduct the experimental comparison among our proposed weight-adjusting approach on the heterogeneous forecasting model, each individual model in the ensemble, and a hybrid homogenous forecasting model (e.g. random forest). It turns out that our proposed approach has a better performance than each single model in the ensemble and the hybrid homogeneous forecasting model in terms of mean absolute error (MAE) and mean absolute percentage error (MAPE).