Forecasting Long-Term Call Traffic Based on Seasonal Dependencies

How to use future call traffic for scheduling different staffs to work in a month or a week is an important task for call center. In this problem setting, the call traffic should be predicted in a long-term way where the forecasting results for different periods are required. However, it is very challenging to solve this problem due to the randomness nature of the call traffic and the multiple forecasting in long term. Current methods cannot solve this problem since they either merely focus on short-term forecasting for the next hour or next day, or ignore call-holding time for call traffic prediction. In this paper, we propose an effective method for predicting long-term call traffic with multiple forecasting results for different future periods, e.g., every 15 min, and take both call arrival rate and call-holding time into consideration through the Erlang. In our method, the seasonal dependencies are summarized by performing data analysis, then different features based on these dependencies are extracted for training the prediction model.In order to forecast call traffic of multiple time buckets, we propose two strategies based on different features. The evaluation results show that the features, the prediction models and the strategies are feasible.