Title: Robust regression-based functional principal component analysis
Authors: Haolun Shi - Simon Fraser University (Canada) [presenting]
Abstract: It is of great interest to conduct robust functional principal component analysis (FPCA) that can identify the major modes of variation in the stochastic process with the presence of outliers. A new robust FPCA method is proposed in a new regression framework. An M-estimator for the functional principal components is developed based on the Hubers loss by iteratively fitting the residuals from the KarhunenLovve expansion for the stochastic process under the robust regression framework. Our method can naturally accommodate sparse and irregularly-sampled data. When the functional data have outliers, our method is shown to render stable and robust estimates of the functional principal components; when the functional data have no outliers, we show via simulation studies that the performance of our approach is similar to that of the conventional FPCA method. The proposed robust FPCA method is demonstrated by analyzing the Hawaii ocean oxygen data and the kidney glomerular filtration rates for patients after renal transplantation.