Abstract: Propose a new agglomerative hierarchical clustering based method to eliminate outliers，with clustering tree to identify outliers.After removing the outliers，build a dynamic prediction model by RBF network，and the experimental results show that the training and generalization performance are markedly improved，which means the agglomerative hierarchical clustering method is effective and workable for outlier detection.After the elimination of outliers，the model shows faster converging speed and higher generalization ability.

Key words: outlier detection, agglomerative hierarchical clustering, radial basis function neural network, prediction

摘要： 提出一种基于凝聚层次聚类消除孤立点的新方法，借助聚类树识别孤立点。去除孤立点后，利用RBF网络建立动态预测模型，实验结果表明，网络的训练和泛化性能较消除孤立点前有明显提高。说明凝聚层次聚类方法用在孤立点检测方面是有效可行的，消除孤立点后建立的模型收敛速度快，泛化能力更优。

关键词: 孤立点检测, 凝聚层次聚类, 径向基神经网络, 预测