Long-Term and Multi-Step Ahead Call Traffic Forecasting with Temporal Features Mining

Abstract

An accurate call traffic forecasting can help the call center to schedule and manage its employees more scientifically. Meanwhile, to meet the needs that some tasks in the call center require the prediction of call traffic in different time buckets for a future long term, it is necessary to forecast the call traffic in a long-term and multi-step way. However, existing forecasting methods cannot solve this problem as (1) Most existing methods merely focus on short-term forecasting for the next hour or the next day. (2) The temporal features of call traffic are ignored, which leads to a lower accuracy in long-term forecasting. Hence, in this paper, we propose a holistic solution for forecasting long-term multi-step ahead call traffic. In our method, we give a categorized way for temporal features by studying the call traffic data. After data preprocessing, we develop an extraction method for temporal features extraction for training the forecasting model. We propose two forecasting strategies based on taking different types of features as input. Experimental results on the real-world call traffic dataset show the effectiveness of our solution, including data preprocessing, temporal features mining, and the forecasting model.