A non-parametric affine-invariant extrapolation depth based classifier that is resistant to extreme values and emissions is proposed and analyzed. A multilevel smoothing structure is proposed that allows obtaining global properties of density functions and class boundaries under appropriate conditions of regularity. Depth extrapolation based classifier uses kernel density estimates for efficient classification of multidimensional data at different levels of smoothing. Refs: 9 titles.
Галкин Александр Анатольевич, кандидат физ.-мат. наук, ассистент кафедры Киевского национального университета имени Тараса Шевченко,
e-mail: galkin.o.a@gmail.com