Comparing RCDU and Ga Statistics for Patch Of Outliers Detection in Von Mises (VM) Samples

Abstract

This paper discovers a new comparison between two outlier detection methods, the Ga statistic and the RCDU statistic, for the identification of outliers in a patch in univariate circular data. This has not been studied in the existing literature before. The RCDU statistic, being an extension from the circular median, is compared to the Ga statistic constructed using the spacing theory. We compare the detection performance of two, three, and four outliers in a patch. The proportion of detecting a correct outlier is used to measure the performance for both test statistics by Monte Carlo simulation. The result shows that the RCDU statistic slightly outperforms the Ga statistic in most cases. This finding aligns with a practical example using coastal wind direction data.