A simple nominal classification algorithm based on qualitative features

An alternative approach to classifying is proposed based on nominal qualitative features. This approach differs from conventional approaches in that instead of comparing test object features tuple with similar features tuples of training sample objects, the independent pairwise comparison is performed for each pair of values in corresponding tuples of objects features. This allows to form the matrix of features weights for each test object that is more detailed than the test object nearest neighborhood. In this approach the simple classification algorithm is suggested, that has a number of important features in respect of classification results interpretation. The quality of the algorithm is tested on an imbalanced sample taken from the known UCI repository. It is shown that the algorithm provides good objects classification accuracy for “small” classes.