A Comparison of Time Series Forecasting Models between Hybrid Model and Individual Model for Forecasting Daily Incoming Call Volume

Abstract

The research aimed to compare the prediction accuracy of forecasting methods for daily incoming call volume. There were three forecasting methods that were used in the study: Box-Jenkins method with SRIMA model and SARIMAX model, Artificial Neural Network (ANN) model and the hybrid model combining SARIMAX and Artificial Neural Network (SARIMAX-ANN) model. The data used in this study is time series of daily incoming call volume to call center which can be divided into 2 data sets. The first data set which was the past data from January 2016 to December 2018 were used for selecting of the most suitable model and the second data set was the past data from January 2019 to December 2019 for the comparison of the accuracy of forecasting model by using Mean Absolute Percentage Error (MAPE). The results showed model with the lowest MAPE is hybrid model of SARIMAX-ANN (MAPE = 24.02%), while the MAPE values for SARIMAX ANN and SARIMA were 24.06%, 43.70%, and 44.97% respectively. It indicates that the hybrid model is more accurate in forecasting than individual model. The hybrid model can be used to forecast daily incoming call volume which is supporting information for the suitable workforce planning of customer service center in the future.