Using Clustering Methods in Multinomial Logit Model for Departure Time Choice

Table 2

Summary of MNL modelling criteria on choice sets from various clustering methods.

Criteria	Clustering method
	K-medoids	E.M.	K-means
Log likelihood at convergence ()	−7700.04	−6853.41	−6166.78
Log likelihood at constant ()	−8753.17	−8073.84	−7259.38
Log likelihood at zero ()	−8905.19	−9171.18	−8975.74
	0.132	0.250	0.310
AIC	1.771	1.576	1.419
Prediction ()	45.73%	52.34%	58.63%