B0667
Title: Enhancing outlier detection in functional data via robustly adjusted functional boxplot
Authors: Andrea Cappozzo - Politecnico di Milano (Italy) [presenting]
Francesca Ieva - Politecnico di Milano (Italy)
Annachiara Rossi - Politecnico di Milano (Italy)
Abstract: Detecting outliers in functional data analysis (FDA) is crucial due to the potential impact of unusual patterns on inference. However, identifying these anomalous curves can be challenging due to the infinite-dimensional nature of such samples. To address this issue, adjusting the fence inflation factor in the functional boxplot is proposed, a widely used tool in the FDA, using simulation-based techniques. This adjustment involves controlling the proportion of observations considered anomalous in a population without outliers, generated through simulation from the original dataset. Robust estimators of location and scatter are required to accomplish this. The effectiveness of high-dimensional multivariate procedures and functional operators in implementing this tuning process is compared. The validity of the proposal is demonstrated through a real example involving the identification of patients with cardiac pathology by means of ECG signals.
