Improved FCM Clustering Algorithm Based on Learning Automata and its Application
Fenglian Li, Zeyi Guo, Xueying Zhang, Jinyi Zhang, Fengyun Hu, Wenhui Jia · 2023
Fuzzy C-means (FCM) clustering algorithm mainly calculates the membership degree of each sample to determine the cluster to which the sample belongs. However, it has the defects of over-reliance on the initial clustering center and not fully considering the influence of the membership matrix change trend on the clustering performance, resulting in poor clustering performance. Aiming at the problems existing in FCM, an improved FCM clustering algorithm based on learning automata is proposed. The proposed algorithm uses an agent to assign a class cluster to each sample, and the probability of each sample being assigned to each class cluster is set equally on initialization. The Q-table of learning automata is composed of the probability of the class cluster to which the sample belongs. On this basis, a new objective function is established, which is associated with the Q-table. The algorithm introduces the variable of average in-class distance, designs rewards function based on the agent actor selection according to the objective function and the change of the average in-class distance to update the Q table. The stopping condition is judged whether the difference between the old and new objective functions is less than the threshold or reaches a certain number of iterations. In the experiments, eight UCI public data sets and the stroke screening data are used to evaluate the effectiveness of the algorithm. Experimental results show that, compared with several existing clustering algorithms such as K-means, FCM, IEWLFCM and LAC, the improved FCM clustering algorithm based on learning automata proposed in this paper has improved accuracy and other clustering indicators for most data sets.