Statistical parameters, like variance and mean value, form the basis of statistical outlier detection methods; these parameters are very sensitive to the presence of outliers in data, and statistical methods inherit this sensitivity from statistical parameters. when an extremely far outlier having a very big or very small magnitude exists in a data set, it usually biases the average and variance with a relatively large magnitude of deviation. This deviation may cause the method to work quiet unexpectedly and loose its functionality. Avoiding these unwanted effects; robustification is of great importance in statistical methods. An outlier detection method, capable of preserving its functionality in the case of presence of outliers is called a robust method. Some robust methods use a robust estimation of statistical parameters instead of the original parameters to attenuate effects of far outliers on the parameters values. In this paper we introduce a robustified outlying detection method based on projection pursuit and robustified using Donoho-Stahel estimators. In this method, we define an outlyingness factor and use it to extract outliers from a data set. We also compare the results in the cases of using robust estimated parameters with the case of using ordinary parameters.
展开▼
