Классификация последовательностей с использованием скрытых марковских моделей в условиях неточного задания их структуры
Гультяева Татьяна Александровна, Попов Александр Александрович · Вестник Томского государственного университета. Управление, вычислительная техника и информатика · 2013
The problem of sequences classification with use of methodology of hidden Markov models (HMM) is considered. Classification is spent both with use of the standard approach, and with classifier of k nearest neighbors (kNN), and a support vector machines in space of the signs initiated by HMM. The behavior of qualifiers was investigated while errors in the structure specification of HMM are presented. Researches have shown that under conditions of structural uncertainty use of qualifiers of k nearest neighbors and the classifier based on a SVM, leads to improvement of classification quality in comparison with the traditional approach based on the ratio of logarithms of likelihood function. Thus the gain in correct classification in the considered two-class problem in comparison with the traditional approach in some cases can reach 40 % both for kNN, and for SVM. The last shows more exact results in comparison with kNN. The maximum improvement reaches about 18 %.
Cited by 0 works.