Design And Implementation Of A Call Center Workforce Forecasting System

In recent years, the cost of labor is becoming more and more expensive. Thereforethe workforce management system is also more and more important, especially for acall center, whose labor cost can count over65%of total cost. Within workforcemanagement system, the forecasting system can efficiently raise the profit-operatorratio, and decrease the idle time of operators, thus, the system turns out to beparticularly significant.This paper describes the design and implementation of a workforce forecastingsystem, based on Support Vector Machine (SVM). Briefly, the system optimizesrelevant parameters of SVM according to the provided data and generates forecastingresults of different KPIs. Then it will use the forecasted data to generate a workforceallocation plan for managers of the call center. Besides, it also enables the user tocustomize these results manually.This paper mainly makes two essential contributions: the first one is the way toimprove the forecasting accuracy. As a solution, a double forecasting strategy isadopted, one is to directly forecast some KPIs (for instances, traffic, average talkingtime) based on historical data, which stands for ‘thinking of the computer’; anotherone is to forecast the user behavior based on his/her previous behavior and historicalgaps between forecasting and actual data, which stands for ‘learning from userbehavior/experience’. Finally, the system can gradually increases the accuracy bycombining these two forecasting results.Second contribution is the implementation of an automatic optimized SVM.According to two observations:1) The superior parameters achieved by the traditionalempirical method cannot meet the requirement of accuracy,2) the optimal parameterusually fall in the neighborhood of these superior parameters upon experiments. Ahybrid algorithm of Particle Swarm Optimization and Generic Algorithm (PSO-GA) isimplemented, which searches the optimal parameters within the neighborhood of thesesuperior parameters. And the result reached a satisfied accuracy and efficiency.This paper presents SVM based workforce forecasting system according to softwarelifecycle in terms of analysis, design, implementation, testing and deployment.Moreover, the detailed design and implementation of key techniques such as interquartile range method, isolation forest, time series features creator, parameteroptimization and PSO-GA is given in the chapter4.