Improving Call Center Load Prediction using News

This thesis evaluates the applicability of news text mining for enhancing daily call center load forecasting for a Dutch flight holiday company called Vliegtickets. nl. A seasonal autoregressive integrated moving average model (SARIMA) is enhanced using news events, that form the external regressors as categorycounts before each predicted day based on the dictionary lists of the General Inquirer (GI). Ant colony optimization is used for feature selection. The used dataset contains the daily number of calls of 2009, 2010, and 2011. While news-based regressors do not consistently improve prediction accuracy in terms of RMSE or MAPE, they improve the RMSE slightly for 2010 that contains extreme news-related peaks.