A network-based discrete choice model for decision-based design

Abstract Customer preference modelling has been widely used to aid engineering design decisions on the selection and configuration of attributes. Recently, network analysis approaches, such as exponential random graph model (ERGM), have increasingly in this field. While ERGM-based approach new capability effects interactions interdependencies (e.g., social relationships among customers) customers’ via structures using triangles peer influence), existing research can only consideration decisions, it cannot predict individual customer’s choices, what traditional utility-based discrete choice models (DCMs) do. However, ability make predictions is essential predicting market demand, which forms basis decision-based (DBD). This paper fills gap by developing a novel for prediction. first time that network-based explicitly compute probability an alternative being chosen from set. Using large-scale customer-revealed database, studies customer preferences estimated with without evaluates their predictive performance benchmarking multinomial logit (MNL) model, DCM. The results show proposed achieves higher accuracy both behaviours share ranking than MNL mathematically equivalent ERGM when no are included. insights obtained study further extend DBD framework allowing explicit entities (i.e., customers products) representations.