Bayesian networks are commonly used for classification: a structural learning algorithm determines the network graph, while standard approaches estimate the model parameters from data. Yet, with few data the corresponding assessments can be unreliable. To gain robustness in this phase, we consider a likelihood-based learning approach, which takes all the model quantifications whose likelihood exceeds a given threshold. A new classification algorithm based on this approach is presented. Notably, this is a credal classifier, i.e., more than a single class can be returned in output. This is the case when the Bayesian networks consistent with the threshold constraint assign different class labels to a test instance. This is the first classifier of this kind for general topologies. Experiments show how this approach provide the desired robustness. 1

Abstract. We propose the COMP-AODE classifier, which adopts the compression-based approach [1] to average the posterior probabilities computed by different non-naive classifiers (SPODEs). COMP-AODE improves classification performance over the wellknown AODE [10] model. COMP-AODE assumes a uniform prior over the SPODEs; we then develop the credal classifier COMP-AODE*, substituting the uniform prior by a set of priors. COMP-AODE * returns more classes when the classification is priordependent, namely if the most probable class varies with the prior adopted over the SPODEs. COMP-AODE * achieves higher classification utility than both COMP-AODE and AODE. 1