ВЫБОР ИНФОРМАТИВНЫХ ПРИЗНАКОВ ДЛЯ ДИАГНОСТИКИ ЗАБОЛЕВАНИЙ ПО ГЕНЕТИЧЕСКИМ ДАННЫМ

In tasks of modern biology, the numbers of attributes often exceed the numbers of objects by orders of magnitude. For the solution of such tasks, a Data Mining method based on using a new measure of similarity between objects in the form of the Function of Rival Similarity (FRiS) is offered. On this basis, methods of quantitative estimation of compactness of patterns, construction of decision rules, and feature selection are developed. All these techniques are implemented in the FRiS-GRAD algorithm. The high efficiency of the algorithm is illustrated by results of solving the task of disease recognition on a microarray dataset.