Home // International Journal On Advances in Intelligent Systems, volume 9, numbers 1 and 2, 2016 // View article

Type of Stochastic Dependence and its Impact on the Performance of Regression Type Classifiers

Authors:
Olgierd Hryniewicz

Keywords: Binary classification; Regression type classifiers; Copulas; Simulation of dependent data; Robustness

Abstract:
Six regression type binary classifiers based on linear and logistic models have been evaluated using a complex simulation experiment. The classifiers were compared with respect to the robustness to unexpected changes of the models that describe data in training and test sets. The data used for this comparison were generated using different models describing their interdependence. This dependence was modeled by different copulas. The experiments revealed that the performance of considered classifiers strongly depends upon the type of copula. However, the simple logistic regression has appeared to be the best one in these circumstances. Thus, this classifier could be recommended for practitioners when the type of dependence may vary in time.

Pages: 161 to 170

Copyright: Copyright (c) to authors, 2016. Used with permission.

Publication date: June 30, 2016

Published in: journal

ISSN: 1942-2679