Classifier Committee Based on Feature Selection Method for Obstructive Nephropathy Diagnosis

摘要

The article presents a multiple classifiers approach to the obstructive nephropathy recognition - a disease posing a significant threat to newborns. Nature of the data reflects a problem known as high dimensionality small sample size. In presented approach a feature space division amongst number of classifiers is used to balance the relation between the number of objects and the number of features. Methods of feature selection are apllied for optimum splitting the feature space for classifier ensemble. The optimal size of subspaces and selection of classifier for ensemble is thoroughly tested. Complex performance test are presented to highlight the most efRcent tuning of parameters for the presented approach, which is then compared to the classical solutions in this field.