Empowering classification for multivariate functional data with simultaneous feature selection.

Area of Science:

• Machine Learning
• Statistical Modeling
• Medical Imaging Analysis

Background:

• The increasing availability of complex imaging data necessitates advanced classification methods.
• Existing functional data classifiers struggle with high-dimensional, multivariate functional predictors and their inter-correlations.

Purpose of the Study:

• To develop a novel classifier for multivariate functional data, particularly for applications like Alzheimer's disease research.
• To address the challenges of high dimensionality and complex correlations in functional data classification.
• To perform simultaneous feature selection and classification.

Main Methods:

• A sparse deep rectified linear unit network architecture was employed.
• The LassoNet algorithm was integrated for feature selection and classification.
• A functional Bayesian information criterion deep neural network was proposed, handling infinite-dimensional predictors.

Main Results:

• The proposed method demonstrated effective feature selection and classification simultaneously.
• The classifier successfully managed complex inter-correlation structures among multiple functional processes.
• Performance was validated through a simulation study and a real-world data application.

Conclusions:

• The novel functional deep neural network classifier offers a powerful approach for multivariate functional data analysis.
• This method provides a robust solution for classification tasks involving complex, high-dimensional imaging data.
• The approach advances the field of statistical classification for neuroimaging and other functional data applications.