基于聚类优化的RBF神经网络多标记学习算法

多标记学习采用RBF神经网络与K-means聚类算法相结合取得了较好的效果,但由于聚类数事先不能很好地确定,无法给出准确的聚类个数值,会导致聚类质量下降、聚类结果不稳定等,进而影响RBF神经网络多标记算法的稳定性及分类性能。本文从样本几何结构的角度出发,采用一种聚类有效性指标函数,为每个类寻找最优的聚类个数,从而优化问题的求解。理论研究和实验结果表明,改进后的算法在分类的稳定性及分类性能方面都有较好的表现。

Multi-label RBF neural networks learning algorithm based on clustering optimization

Multi-label learning,combining RBF neural network and K-means clustering algorithm,has achieved good effects.But because the number of clusters cannot be well determined in advance,an accurate value of the clustering cannot be obtained.This problem will lead to lower quality clustering and clustering instability,and then affect the stability and the classification performance of the multi-label RBF neural network algorithm.To solve the optimization problems,from the angle of sample geometry,an index function for clustering validity was employed to find the optimal number of clusters for each class.Theoretical research and experimental results show that the improved ML-IRBF algorithm can effectively boost better performance in terms of the stability and capability of classification.