A1318
Title: Anomaly detection for functional data
Authors: Hyemin Yeon - Kent State University (United States) [presenting]
Xiongtao Dai - University of California, Berkeley (United States)
Sara Lopez Pintado - Northeastern University (United States)
Abstract: Anomaly detection is a fundamental task in Statistics and Machine Learning, especially as an initial step in exploratory data analysis. Techniques such as boxplot for univariate data, bagplot for multivariate data, and functional boxplot for functional data have been widely used to identify outliers across various domains. The focus is on functional data, introducing a new approach for detecting complex outliers using a novel concept of halfspace depth. This method is particularly effective in identifying functional shape outliers, which are known to be difficult to detect with traditional techniques. Numerical studies show that the proposed method outperforms existing approaches in recognizing various types of functional anomalies. Its practicality is demonstrated through applications to real-world datasets.
