Combining biased regression with machine learning to conduct supply chain forecasting and analytics for printing circuit board

Demand planning (DP) and sales forecasting (SF) are classical, yet, important techniques in supply chain management. In practice, demand planning is usually treated as the optimisation of internal resources (mathematical programming and game-theoretic modelling) while sales forecasting is related to the impacts of external markets (statistical regression, time series, machine learning). To help printing-circuit-board (PCB) manufacturers conduct effective DP and SF, a demand-driven framework is presented to accomplish the following goals. First, least absolute shrinkage and selection operator (LASSO) is employed to identify significant consumer products (smartphone, tablet, panel display, desktop, laptop, server, printer) affecting PCB sales volumes and revenues. Second, Holt-Winter’s smoothing (HWS) is constructed to predict global shipments of consumer products. Finally, statistical regression and ensemble learning are adopted to conduct forecasting and sensitivity analyses. In summary, the presented framework helps PCB manufacturers incorporate seasonal variations of consumer products into demand planning for PCB shipments and sales forecasting for PCB revenues. Sensitivity analyses identify panel display and laptop significantly as key products for PCB shipments while smartphone and server are critical to PCB revenues.