1. 연사 : 박민수 교수 (충남대학교 정보통계학과)

2. 주제 : Adaptive Fences for Robust Outlier Detection under Skewed Distributions

3. 일시 : 2025년 11월 27일(목) 16:00

4. 장소 : 초청 강의실 (자1-124호)

5. 발표 초록 :  

Outlier detection is an essential element of statistical analysis as it ensures the integrity of data and the validity of resulting inferences. Classical approaches such as Tukey’s boxplot are straightforward and widely used, yet they often overidentify outliers when the underlying distribution is skewed. Hubert and Vandervieren (2008) proposed the adjusted boxplot that incorporates a robust skewness measure based on the medcouple, thereby improving detection in skewed settings. However, the computational cost of this approach can be substantial, which restricts its use in large datasets or in contexts where repeated analysis is required.

This study develops an outlier detection method that modifies the boxplot framework by employing a skewness-adjusted fence based on a measure derived from the median absolute deviation. The method is designed to retain statistical reliability while improving computational efficiency relative to existing approaches. Results from simulation experiments and applications to real data indicate that the proposed procedure provides more accurate detection under skewed distributions and performs well in terms of efficiency. The method is further extended to time-dependent data and is shown to be effective in dynamic settings, with potential applications in areas such as finance, healthcare, and environmental monitoring where reliable anomaly detection is required.

Keywords: Robust outlier detection, Skewness-adjusted boxplot, Influence function, Median absolute deviation