B0905
Title: Robust parameter estimation, variable selection in the ultra-high dimensional regression with autoregressive error terms
Authors: Yetkin Tuac - Ankara University (Turkey) [presenting]
Peter Filzmoser - Vienna University of Technology (Austria)
Olcay Arslan - Ankara University (Turkey)
Abstract: Due to its undeniably expanding field of use, many studies have been carried out for the analysis of ultra-high dimensional data from various perspectives. The aim is to develop a method for the analysis of ultra-high dimensional data considering the presence of outliers and autocorrelation structure in the error terms. First, screening methods are employed to reduce the high dimensionality from p to d (d<n). After reducing the dataset to dimension d, penalty-based methods are applied in combination with robust techniques to achieve simultaneous parameter estimation and variable selection. The behavior of the method is illustrated through a simulation study conducted under different scenarios and its superiority is demonstrated over existing methods when both outliers and autoregressive structures exist in the dataset. Additionally, a real data example is provided to support the idea.
