Predicting ski lodging selection using customer databases

This research illustrates the advantages that the application of data-mining techniques can bring into designing a direct marketing campaign. Specifically, it presents results of two multinomial logit models. The models identified patterns in the data and established the relationship between characteristics of repeat visitors and their lodging preferences. Based on this relationship it is possible to predict what type of accommodation customers would prefer at a level better than chance and to customize marketing messages to meet customers’ needs. Two of the strongest predictors of choic e are income and length of stay for the previous visit. The project also offers lift chart analysis of the multinomial models’ results that led to better understanding of the two models’ predictive powers, a visualization tool rarely used for interpreting multinomial models. This analysis revealed that the model identified 94% of all visitors that prefer Silver accommodation within 50% of the top scoring customers.