Augmented naive Bayesian model of classification learning

Principles of machine learning are used to propose a unified model of human categorization. The model is a memory constrained approximation of the optimal classifier. The model augments the Naive Bayesian Classifier with feature construction using a Galois lattice and feature selection guided by feature worthiness. The classifier unifies a set of human categorization phenomena under the principles of optimality, utility of feature worthiness and smaller representations for categorization. The Naive Bayesian Classifier and an Augmented Naive Bayesian Classifier (ANB) are fit to ten human classification data sets. The results show the ANB model predicts more of the variance in classification phenomena and supplies an account for organizing the space of classification behaviors.