A Novel Genetic Algorithm Based ECOC Algorithm

This paper proposes a genetic algorithm (GA) based error correcting output codes (ECOC) algorithms. In our algorithm, some randomly initialized coding matrices are generated as seeds firstly, and our algorithm produces optimal coding matrices based on them in the evolutionary process. In our GA, each gene stands for an action, indicating two selected columns and an operator. The operators are proposed to generate new columns by exchanging information between a pair of parent columns. In this way, each individual represents a new coding matrix. A legality checking function is embedded in the GA to keep the produced coding matrix both legal and effective. At the end of this evolutionary process, the best coding matrix is selected as final solution. The experimental results show that our algorithm can efficiently optimize the coding matrix compared with the seed matrices.