Abstract
We have introduced in this paper the concept of structural hidden Markov models (SHMM's). This new paradigm adds the syntactical (or structural) component to the traditional HMM's. SHMM's introduce relationships between the visible observations of a sequence. These observations are related because they are viewed as evidences of a same conclusion in a rule of inference. We have applied this novel concept to predict customer's preferences for automotive designs. SHMMhas outperformed both the k-nearest neighbors and the neural network classifiers with an additional 12% increase in accuracy.