Derived Multi-population Genetic Algorithm for Adaptive Fuzzy C-Means Clustering.
Weiping Ding,Zhihao Feng,Javier Andreu-Perez,Witold Pedrycz
DOI: https://doi.org/10.1007/s11063-022-10876-9
IF: 2.565
2022-01-01
Neural Processing Letters
Abstract:Fuzzy C-Means (FCM) is a common data analysis method, but the clustering effect of this algorithm is easily affected by the initial clustering centers. Currently, scholars often use the multiple population genetic algorithm (MPGA) to optimize the clustering centers, but the MPGA has insufficient global search ability and lacks self-adaptability, is prone to premature convergence, and has poor initial clustering centers. Therefore, this paper proposes an adaptive FCM clustering algorithm DMGA-FCM based on a derivative multiple population genetic algorithm (DMGA). In DMGA-FCM algorithm, firstly, the derivative operator, which is proposed for the first time in this paper, performs derivative operations on initialized populations to improve the algorithm's searchability and deal with the lack of inter-population search ability. Secondly, the adaptive probability fuzzy control operator is used to dynamically adjust the genetic probability to improve the adaptability of the algorithm, which in turn enhances the global merit-seeking ability of the DMGA algorithm and avoids premature convergence. Finally, the initial clustering center of FCM algorithm is optimized with DMGA to enhance the clustering effect of the algorithm. The analysis of simulation experiments and MRI brain map application results show that the DMGA-FCM algorithm can obtain a better clustering effect of medical data and image clustering segmentation effect compared with other related FCM algorithms.