Гібридний алгоритм кластер-аналізу для формування апріорного розбиття простору ознак на класи знань в системах дистанційного навчання

Abstract

There has been offered the modification of algorithm of k - means, the idea of improvement of which consists in the combined use of criterion of estimation of error of clusterization and informative criterion of functional efficiency, that determines authenticity of the built decision rules of determination of belonging of realization to some class of knowledge. Thus the simultaneous use of statistical and informative approaches allowed including such important parameter for the algorithms of clusterization as an amount of clusters in iterative optimization procedure. Having a priori information about distribution of N- measure vectors of realization, presenting the results of testing of knowledge of students, it also allows to define the optimal geometrical parameters of containers, describing the classes of knowledge of students in the systems controlled from distance education.