Attribute ranking for intelligent data analysis in medical applications

The work critically analyzes machine learning based attribute selection algorithms from the perspective of their applicability for intelligent data analysis. Different approaches are illustrated by the results obtained by their application on a large medical domain of heart failure patients. Random Forest algorithm, based on voting of many relatively non-correlated classifiers, is accepted as the most reliable approach to attribute ranking. Additionally, it is demonstrated that rule-based machine learning algorithms can be used for feature ranking and that application of rule quality measures with different generality may be very useful for human understanding of the domain.