Multivariate outlier detection: A comparison among two clustering techniques

Abstract

In data analysis, the first step for broad analysis is outlying items. Outlying items cause false results, biased parameter estimation and model misspecification. Some of the methods for detection of outlier are: Distance based method, Distribution based method, Density based method and Clustering based method. In this research, clustering based method is used for outlying items. The purpose of this study is to detect outlier in multivariate data by performing cluster analysis. The K means and Partitioning Around Medoid (PAM) methods were performed. After cluster the data outliers are detected by measuring distance. The comparison of the clustering techniques in outlier detection methods are analyzed. The secondary data was used for analysis. In this data set impact of zinc phosphide with three sub lethal doses (concentrations) on different body tissues of rat are used to find any anomalies. R software was used for data analysis.