Design and Implementation of a Bayesian Classifier with Automatic Complexity Regulation

A growing demand exists for sophisticated algorithms that can learn predictive models from data for a wide range of engineering and scientific applications. In this paper, we introduce a novel, parameter-free method for constructing Bayesian Network Augmented Naive Bayes classifiers directly from data. Our approach dynamically adjusts model complexity to the available training data, thereby mitigating the risk of overfitting. We evaluate the proposed algorithm on ten UCI Machine Learning Repository datasets, demonstrating that it consistently matches or exceeds the predictive accuracy of Tree-Augmented Naive Bayes (TAN) and remains competitive with three leading state-of-the-art classifiers.