An algorithm for competitive learning in clustering problems

摘要

Competitive learning is one of the algorithms for clustering based on the least sum of squares criterion. However, competitive learning has a problem of serious decline of learning ability from lack of competition when one unit monopolizes all input vectors. This problem must be avoided using an additional algorithm. This paper presents an algorithm which makes competitive learning give a good approximate solution for clustering. The results using our method are better than previous studies. This paper also presents a parameter optimization of competitive learning using ANOVA (analysis of variance). Using ANOVA helps to optimize a parameter condition systematically.