Using Bayesian belief network and time-series model to conduct prescriptive and predictive analytics for computer industries

Abstract

Business intelligence & analytics (BI&A) has become an important area for both researchers and practitioners. The conventional business intelligence emphasizes descriptive and diagnostic analytics to achieve performance measurement and management. Furthermore, business analytics extends to include predictive and prescriptive analytics to generate responsive action plans. In the area of BI&A, the following issues are critical but difficult to tackle: (1) How to identify key performance indicators (KPIs) in a data-driven manner? (2) How to consider the interrelationships among the KPIs to develop a strategic map for an outcome? (3) How to incorporate the impacts of the leading indicators on a lagging outcome into performance forecasting? Inspired by the concept of statistical learning and machine learning, this research presents a novel framework consisting of random forest, Bayesian belief network, and time-series model. In particular, to justify the validity of the presented framework, three types of personal-computer (PC) firms including commercial PC, industrial PC and fabrication PC are respectively used to characterize various business models in computer industries: original brand manufacturing (OBM), original design manufacturing (ODM), and electronic manufacturing service (EMS).