Evaluation of Some Outlier Detection Methods based on Real Life Data Application

Summary

The research work evaluated some outlier detection techniques such as t-statistic (T), Modified Z-Statistic (τ ), Cancer
Outlier Profile Analysis (COPA), Outlier Sum-Statistic (OS), Outlier Robust T-Statistic (ORT), and the Truncated
Outlier Robust T-Statistic (TORT) to verify which technique has the highest power of detecting outliers on the bases of
their Rank Values, P-values, True Positives(Sensitivity), False Positives (Specificity) and False Discovery Rate (FDR)
using real life data application. It was observed using the Rank Values that OS has the highest Rank Value followed by
t-statistic, ORT, TORT, Z while COPA had the least rank. It was also observed using the P-values that COPA
performed better than the other methods by having the highest number of False Positives (specificity) followed by OS
with a better specificity (FP) and sensitivity (TP) while Z, T, ORT and TORT have no False Positive. In terms of their
False Discovery Rate (FDR), the performance of OS is outstanding with a smaller FDR followed by COPA, T, ORT,
TORT and Z.

Index Terms

Browse additional articles authored by one or more contributors to this paper.