An Outlier Detection Algorithm Based on Clustering Analysis

Outlier detection is a hot topic of data mining. After analyzing current detection technologies, a detection method of outlier based on clustering analysis is proposed, in which an effective sample is screened out from original data. According to agglomerative of hierarchical clustering, credible sample set is found. Then mathematical expectation and standard deviation are obtained by credible sample. Finally, global data will detected by the definition of outlier which is proposed in this paper. The data disposed by this method can be irrelative to the time scales. And it needs not to presuppose the number of outlier. The experiment results on IRIS show that this method can detect outliers effectively.