A0567
Title: Outlier detection for high-dimensional data by principal component analysis
Authors: Yugo Nakayama - Nissan Motor Co., Ltd. (Japan) [presenting]
Kazuyoshi Yata - University of Tsukuba (Japan)
Makoto Aoshima - University of Tsukuba (Japan)
Abstract: Outlier detection is examined for high-dimensional, low-sample-size (HDLSS) data. By deriving asymptotic properties of principal component scores with and without outliers, a novel detection method is proposed. Using this approach, an algorithm is developed to identify multiple outliers simultaneously. The theoretical properties of sure independent screening are investigated, and its ability to achieve complete outlier identification with high accuracy is demonstrated. Comparative analyses against existing HDLSS outlier detection methods are conducted through both numerical simulations and real data applications. The proposed method demonstrates superiority not only in correctly identifying true outliers but also in minimizing false identifications. This comprehensive framework provides a robust solution to the challenging problem of outlier detection in high-dimensional contexts where traditional methods often fail.
