Interest/intention classification for the Fish-Bird new media artwork

This paper investigates the appropriateness of various features and classifiers to an interest classification task. The data for the task is obtained from various exhibitions of the Fish-Bird artwork. A number of features from the dataset are calculated and described, based on psychology literature on human motion. These features are used in both a Dynamic Bayesian Network classifier, and a Conditional Random Field classifier, representative types of both Discriminative and Generative classifiers. These two trained classifiers are both discussed along with a Heuristic-based approach. The Classification Accuracy for the two trained classifiers is compared using both real-world data and simulated data.