Reverse Logistics Design with Neural Networks

Background

To the best of our knowledge, the number of RL studies that has a comprehensive forecasting methodology on product return amount is low because of RL’s complicated structure. Kelle and Silver (1989), Hess and Mayhew (1997), Gomez et al. (2002), Toktay et al. (2004), Hanafi et al. (2007), Klausner and Hendrickson (2000), Guide and Van Wassenhove (2001) analyse product return processes. The drawback of these studies is that they are mostly taking a few factors into consideration at forecasting process. In a successful forecasting method, an extended factor set should be taken into consideration. The factors that are compiled from European Union legal regulation on e-wastes (WEEE Directive), different studies in the RL literature and negotiations with managers could be classified into two main categories: macro and micro factors (as seen in Table 1).

On the other side, a quick look at RLND literature reveals that the main topics range from strategic decisions to operational decisions. As Dekker et al. (2004) state, RLND is one of important topics in RL research area. One of the main issues in RL is to define the best locations for WEEE recycling centers (Queiruga et al., 2008). There are many operations research models generated by researchers for obtaining optimal solution under specific objectives and constraints (Spengler et al., 1997; Krikke et al., 1999; Fandel & Stammen, 2004; Listeş, 2007; Sasikumar et al., 2010; etc.). The researchers mostly assume that all inputs and all relationships of the model are known with certainty. In the literature, uncertainty is commonly considered by scenario analysis in the mixed integer network design (Dekker et al., 2004) or by some stochastic programming models but there is lack of hybrid studies regarding forecasting system development.