applied Support Vector Machine. (SVM) into conjoint analysis. SVM is a popular machine learning algorithm generally used in classification problems

[PDF]Multidomain Demand Modeling in Design for Market Systems

https://deepblue.lib.umich.edu/.../nwkang...

University of Michigan Library

by N Kang - ‎2014 - ‎Cited by 4 - ‎Related articles

(2005) applied Support Vector Machine. (SVM) into conjoint analysis. SVM is a popular machine learning algorithm generally used in classification problems.

[PDF]Crowdsourced Design for Product Form and Functions: Bi ...

https://namwookanghomepage.appspot.com/papers/WorkingPaper2.pdf

conjoint analysis. SVM is a popular machine learning algorithm generally used in classification problems. Especially, Evgeniou et al. (2005) proposed a SVM ..

https://deepblue.lib.umich.edu/bitstream/handle/2027.42/110471/nwkang_1.pdf?sequence=1&isAllowed=y

http://cbcl.mit.edu/cbcl/publications/theses/thesis-zacharia.pdf


For demand modeling, Wassenaar and Chen (2003) adopted discrete choice analysis (DCA) assuming that a customer chooses a product whose utility is the highest among given product options. DCA has been widely used as the standard demand modeling technique in marketing and in DMS. To measure utility of each product option, the random utility concept is used, uij = vij + εij , (2.2) where uij is utility that product j provides to individual i, v is a deterministic component that can be observed, and εij is an error component that cannot be observed. In marketing, the deterministic component vij is generally assumed as a linear function of discrete levels of attributes and is defined as vij = X K k=1 X Lk l=1 βiklzjkl, (2.3) where zjkl are binary dummy variables indicating product j possesses attribute k at level l, and βikl are the part-worth coefficients of attribute k at level l for individual i (Green and Krieger , 1996b). The probability that individual i chooses product j from a set of options J can 20 be defined as the probability that product j has a higher utility than all alternatives: Pij = Pr[uij > uij0; ∀j 0 ∈ J] = Pr[vij + εij > vij0 + εij0; ∀j 0 ∈ J]. (2.4) If the error component εij is assumed to be independently and identically distributed (iid) across choice alternatives and if it follows the extreme value distribution7 (Gumbel, Weibull or double exponential), then the probability Pij can be estimated by the multinomial logit (MNL) model as Pij = e vij P j 0∈J e vij0 . (2.5) MNL estimates part-worth coefficients using the maximum likelihood method with consumer preference data. The main limitation of MNL is the Independence of Irrelevant Alternatives (IIA) modeling assumption that the utility of each product alternatives has the same error component, so that the choice probability between two alternatives is not affected by other alternatives. While MNL is the most widely-used demand modeling technique in DMS research as shown in Table 2.1, some DMS studies have used advanced DCA models such as mixed logit (Shiau et al., 2007; Shiau and Michalek, 2009b; Shiau et al., 2009; Morrow et al., 2014b), nested logit (Kumar et al., 2007, 2009), ordered logit (Kumar et al., 2007; Hoyle et al., 2011), and Hierarchical Bayesian (HB) models (Michalek et al., 2006; Hoyle et al., 2010; He et al., 2011; Michalek et al., 2011; Wang et al., 2011b; Kang et al., 2013a, 2014). Kumar et al. (2007) and Hoyle et al. (2010) have integrated different DCA models into a hierarchical structure. MNL is an aggregate logit model assuming that consumer preference is homogeneous so that the part-worth coefficients in Eq. (2.3) are deterministic and the same across individuals. M