|
Abstract:
|
Outlier detection methods for functional data have been evolving over the past decade, but currently these approaches only detect outliers from a set of complete curves. While this procedure is appropriate for identifying outliers to aid in further analysis of the data, there are many situations for which we would like to detect an outlier dynamically or in real-time so we can intervene. For example, in the case of child growth curves, it would be beneficial for a child to be identified as an outlier as soon as possible so that appropriate medical care can be administered. Inspired by this problem, we consider dynamic outlier detection of outlying observations for functional data observed discretely. Our proposed method is based on functional principal component analysis. We illustrate the method numerically through simulations and through a real data application.
|
