MULTI-OBJECTIVE LEARNING AUTOMATA: AN APPROACH FOR DESIGNING BI-OBJECTIVE CLASSIFIER
Seyed Hamid Zahiri · 2010
A novel multi-objective optimizer has been introduced based on the learning automata (LA) and utilized to develop a multi-objective classifier (named MLA- classifier). The proposed classification method is able to approximate the decision hyperplanes in such way that two performance aspects (i.e. score of recognition and precision) are simultaneously maximized. The proposed MLA-classifier passes three designing phases of training, validity estimation, and testing. A validation function has been defined for selecting the best compromise solution (hyperplanes set). Extensive experimental results on different kinds of benchmarks and practical problems with nonlinear, overlapping class boundaries and different feature space dimensions are provided to show the powerfulness of the proposed multi-objective classifier. Experimental results demonstrate that the performances of the proposed classifier are much better than those of the single-objective LA based classifier. Also the comparative results illustrate that the performances of the proposed bi-objective classifier is comparable to, sometimes better than those of the similar approaches which have been designed based on the GA and S-model learning automata.