Showing 1–3 of 3 results for author: Darkhovsky, B

Models with varying structure

Authors: Boris Brodsky, Boris Darkhovsky

Abstract: In this paper the problems of the retrospective analysis of models with time-varying structure are considered. These models include contamination models with randomly switching parameters and multivariate classification models with an arbitrary number of classes. Our main task here is to classify observations with different stochastic generation mechanisms. A new classification method is proposed.… ▽ More In this paper the problems of the retrospective analysis of models with time-varying structure are considered. These models include contamination models with randomly switching parameters and multivariate classification models with an arbitrary number of classes. Our main task here is to classify observations with different stochastic generation mechanisms. A new classification method is proposed. We analyze its properties both theoretically and empirically. The asymptotic optimality of the propodsed method (by the order of convergence to zero of the estimation error) is also established. At the end of the paper we consider multivariate change-in-mean models and multivariate regression models. △ Less

Submitted 30 October, 2017; originally announced October 2017.

Asymptotically Optimal Detection of Changes in Stochastic Models with Switching Regimes

Authors: Boris Brodsky, Boris Darkhovsky

Abstract: This paper deals with the problem of asymptotically optimal detection of changes in regime-switching stochastic models. We need to divide the whole obtained sample of data into several sub-samples with observations belonging to different states of a stochastic models with switching regimes. For this purpose, the idea of reduction to a corresponding change-point detection problem is used. Both univ… ▽ More This paper deals with the problem of asymptotically optimal detection of changes in regime-switching stochastic models. We need to divide the whole obtained sample of data into several sub-samples with observations belonging to different states of a stochastic models with switching regimes. For this purpose, the idea of reduction to a corresponding change-point detection problem is used. Both univariate and multivariate switching models are considered. For the univariate case, we begin with the study of binary mixtures of probabilistic distributions. In theorems 1 and 2 we prove that type 1 and type 2 errors of the proposed method converge to zero exponentially as the sample size tends to infinity. In theorem 3 we prove that the proposed method is asymptotically optimal by the rate of this convergence in the sense that the lower bound in the a priori informational inequality is attained for our method. Several generalizations to the case of multiple univariate mixtures of probabilistic distributions are considered. For the multivariate case, we first study the general problem of classification of the whole array of data into several sub-arrays of observations from different regimes of a multivariate stochastic model with switching states. Then we consider regression models with abnormal observations and switching sets of regression coefficients. Results of a detailed Monte Carlo study of the proposed method for different stochastic models with switching regimes are presented. △ Less

Submitted 24 January, 2013; originally announced January 2013.

Retrospective change-point detection and estimation in multivariate linear models

Authors: Boris Brodsky, Boris Darkhovsky

Abstract: In this paper the problem of retrospective change-point detection and estimation in multivariate linear models is considered. The lower bounds for the error of change-point estimation are proved in different cases (one change-point: deterministic and stochastic predictors, multiple change-points). A new method for retrospective change-point detection and estimation is proposed and its main perform… ▽ More In this paper the problem of retrospective change-point detection and estimation in multivariate linear models is considered. The lower bounds for the error of change-point estimation are proved in different cases (one change-point: deterministic and stochastic predictors, multiple change-points). A new method for retrospective change-point detection and estimation is proposed and its main performance characteristics (type 1 and type 2 errors, the error of estimation) are studied for dependent observations in situations of deterministic and stochastic predictors and unknown change-points. We prove that this method is asymptotically optimal by the order of convergence of change-point estimates to their true values as the sample size tends to infinity. Results of a simulation study of the main performance characteristics of proposed method in comparison with other well known methods of retrospective change-point detection and estimation are presented. △ Less

Submitted 26 October, 2011; originally announced October 2011.