銷售預測在供應鏈管理中佔有相當重要之角色,且資訊產業相較於其他產業而言,變動更加快速。因此若能建構一個精確有用的銷售預測模式,便可增進企業在供應鏈以及銷售管理的能力。本研究透過多種資料探勘(data mining)預測技術,包括倒傳遞類神經網路(back-propagation network, BPN)、支援向量迴歸(support vector regression, SVR)、多元適應性雲形迴歸(multivariate adaptive regression splines, MARS)以及小腦系統控制器(cerebellar model articulation controller, CMAC)等方法對資訊產業做銷售預測及應用。本研究應用某電腦品牌代理商之每月銷售資料為實證資料,評估資料探勘預測技術於電腦代理商銷售預測的有效性。實證結果顯示,在單一模式下,MARS與SVR的預測準確率較佳,BPN與CMAC的預測績效較差;而在混合模式下,則以整合MARS與SVR的預測模式較佳。
Sales forecasting is one of the most important issues in managing computer reseller since computer products are characterized by product variety, rapid specification changes and rapid price declines. In this study, we use four different data mining prediction methods including back-propagation network (BPN), support vector regression (SVR), multivariate adaptive regression splines (MARS) and cerebellar model articulation controller (CMAC) to forecast sales for computer reseller and compare their forecasting performance. Experimental results from a real sales data of computer reseller show that the prediction accuracy of single MARS and single SVR models are the better than that of single BPN and CMAC. The integrated MARS-SVR model outperforms all other four single and two integrated forecasting models and is a promising model for forecast sales for computer reseller.