No abstract

No abstract

We experimentally learn structures of Bayesian networks classifying signals enabling genetic abnormality diagnosis. Structures learned based on the naive Bayesian classifier, expert knowledge or using the K2 algorithm are compared. Inferiority of the K2-based classifier has motivated an investigation of the algorithm initial ordering, search procedure and metric. Replacing the K2 search with hill-climbing search improves accuracy as does the inclusion of hidden variables into the structure. However, it is proved experimentally that this inferiority of the K2-based classifier is mainly due to the K2 metric soliciting structures having enhanced representability but limited classification accuracy.

The structure and parameters of a belief network are learned in order to categorize cytogenetic images enabling the detection of genetic syndromes. We compare a structure learned from the data to another obtained utilizing expert knowledge and to the naive Bayesian classifier. We also study feature quantization needed for parameter learning in comparison to density estimation. Both networks achieve comparable accuracy for the cytogenetic database with a slight advantage to that based on expert knowledge.