A Novel Support Vector Classifier for Longitudinal High-Dimensional Data and Its Application to Neuroimaging Data

• Activity Number: 63
• Type: Topic Contributed
• Date/Time: Sunday, July 31, 2011 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Shuo Chen*+ and DuBois Bowman
• Companies: Emory University and Emory University
• Keywords: statistical learning ; longitudinal high dimensional data ; SVM ; neuroimaging data ; quadratic programming ; optimization

Abstract

Recent technological advances have made it possible for many studies to collect high dimensional data (HDD), such as images, repeatedly over time. Such studies may yield temporal changes in selected features that, when incorporated with machine learning methods, are able to predict clinical outcomes. However, current methods, such as the support vector machine (SVM), for HDD analysis typically consider cross-sectional data collected during one time period. We propose a novel support vector classifier for longitudinal HDD that allows simultaneous estimation of the SVM separating hyperplane parameters and temporal trend parameters, which determine the optimal means to combine the longitudinal data for classification and prediction. Our approach is based on an augmented kernel function in reproducing kernel Hilbert space and uses quadratic programming for optimization. The results of a simulation study and a data example indicate that our proposed method leverages the additional longitudinal information to achieve higher accuracy than methods using only cross-sectional data and methods that naively combine longitudinal data by simply stacking the data to expand the feature space.