Problem Difficulty for Genetic Algorithm in Combinatorial Optimization

Abstract

This paper presents how difficult to handle (genetic algorithm) GA with combinatorial approach in clustering problem and an alternative approach is suggested. Clustering problem can be viewed as combinatorial optimization. In this paper, the objects must be clustered are new students. They must be allocated into a few of classes, so that each class contains students with low gap of intelligence. Initially, we apply GA with combinatorial approach. But experiments only provide a small scale case (200 students and 5 classes). Then we try to apply GA with binary chromosome representation and we evaluate it with the same data. We have successfully improved the performance with this approach. This result seems to indicate that GA is not effective to be applied for solving combinatorial optimization problems in general. We suggest that binary representation approach should be used to avoid this difficulty.