Bayes classification of imprecise information of interval type
Piotr Kulczycki, Piotr Andrzej Kowalski · Control and Cybernetics · 2011
The subject of the investigation presented here is Bayes classification of imprecise multidimensional information of in- terval type by means of patterns defined through precise data, e.g. deterministic or sharp. For this purpose the statistical kernel esti- mators methodology was applied, which makes the resulting algo- rithm independent of the pattern shape. In addition, elements of pattern sets which have insignificant or negative influence on the correctness of classification are eliminated. The concept for realiz- ing the procedure is based on the sensitivity method, used in the domain of artificial neural networks. As a result of this procedure the number of correct classifications and - above all - calculation speed increased significantly. A further growth in quality of classifi- cation was achieved with an algorithm for the correction of classifier parameter values. The results of numerical verification, carried out on pseudorandom and benchmark data, as well as a comparative analysis with other methods of similar conditioning, have validated the concept presented here and its positive features.