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

Area of Science:

• Biomedical data analysis
• Machine learning
• Longitudinal studies

Background:

• High-dimensional data (HDD) collection is increasing in longitudinal studies, offering potential for disease prediction.
• Current Support Vector Machine (SVM) methods often analyze only cross-sectional data, limiting their ability to leverage temporal information.
• Integrating temporal changes from HDD is crucial for accurate disease status and treatment response prediction.

Purpose of the Study:

• To develop a novel Support Vector Classifier (SVC) for longitudinal high-dimensional data (HDD).
• To enable simultaneous estimation of SVM hyperplane and temporal trend parameters for optimal data combination.
• To enhance classification and prediction accuracy by effectively utilizing longitudinal information.

Main Methods:

• Proposed a novel SVC tailored for longitudinal HDD analysis.
• Developed an approach based on an augmented reproducing kernel function.
• Utilized quadratic programming for simultaneous optimization of hyperplane and temporal trend parameters.

Main Results:

• The proposed SVC method demonstrated higher accuracy compared to cross-sectional SVM.
• Effectively leveraged longitudinal information for improved predictive performance.
• Outperformed methods that naively expand the feature space with longitudinal data.

Conclusions:

• The novel SVC effectively integrates longitudinal HDD for enhanced disease prediction.
• This approach offers significant advantages over traditional cross-sectional methods.
• The methodology shows promise for applications in areas like Alzheimer's disease research.