Rough Neural Network Ensemble for Interval Data Classification
Robert K. Nowicki, Marcin Korytkowski, Rafał Scherer · 2018
The ensemble of rough neural networks designed for the classification of objects described by vectors of intervals is proposed. In contrast to imputation, marginalisation and similar solutions for dealing with imperfect data, the proposed architecture is able to provide an imprecise answer as a result of input data imperfection. It is achieved thanks to incorporating the rough set theory to feedforward neural networks. The AdaBoost algorithm is adapted as a meta-learning of the rough ensemble. We describe gradient training for member rough networks. The proposed architecture and training methodology was tested on a wide range of interval data.