Research on Forecasting Call Center Traffic through PCA and BP Artificial Neural Network

Accurately forecasting future call volumes is critical for scheduling of a call center. This thesis develops a PCA-BP model to forecast future call volumes of half-hour periods. the approach adopted firstly uses principle component analysis to eliminate the intraday correlations between the call volumes of 48 consecutive half-hour periods and to simplify the structure of BP neural network by dimension reduction. the processed data are then input into BP network for training. We use the trained network to forecast future call volumes and apply a competing model to the same data. It turns out that the new model performs better and can be adapted in call center traffic forecasting. to the best of our knowledge, the forecasting method we built has not been used in this area hitherto and it deserves trial application accordingly.