Maintaining Individual Diversity by Fuzzy c -Means Selection

In a GA search, maintaining diversity of individuals is an effective approach for preventing premature convergence and finding multiple optima. Our research aims to maintain the diversity. In this paper, a new selection for maintaining the diversity is proposed, and the selection is applied to simple GA (sGA). In the selection, the individuals are classified by Fuzzy c -means (FCM). Accordingly, several clusters are identified and each of the individuals gets a membership value for each of the clusters. The proposed selection selects individuals based on both the fitness values and the membership values. We discuss the behavior of maintaining individual diversity and search capabilities of the GA with the proposed selection via comparative experiments with a crisp cluster-based selection. Based on the results of the experiments, we were able to determine that the GA with the proposed selection makes the individuals wider distributed in a solution space compared to the crisp clustering based selection. The GA were also able to find more applicable optima compared to sGA and GA with a crisp clustering selection.