Robust classification of functional and quantitative image data using functional mixed models.

Area of Science:

• Statistics
• Machine Learning
• Biostatistics

Background:

• Functional data analysis deals with complex, high-dimensional data.
• Classification of such data presents significant statistical challenges.
• Existing methods may struggle with outliers and complex functional dependencies.

Purpose of the Study:

• To introduce novel classification methods for high-dimensional functional data.
• To extend the functional mixed model (FMM) framework for classification tasks.
• To develop Bayesian approaches for robust functional data classification.

Main Methods:

• Utilizing the functional mixed model (FMM) framework with functional fixed and random effects.
• Implementing Bayesian schemes for training and prediction, incorporating class designation as a fixed effect.
• Developing Gaussian, wavelet-based FMM (G-WFMM) and robust, wavelet-based FMM (R-WFMM) models.
• Employing wavelet-based modeling for parsimonious representation and adaptation to local features.

Main Results:

• The proposed FMM methods effectively classify complex, high-dimensional functional data.
• Wavelet-based approaches provide efficient and adaptive function representations.
• The robust FMM (R-WFMM) demonstrates resilience to outliers in functional data.
• The methods were successfully applied to a pancreatic cancer mass spectroscopy dataset.

Conclusions:

• The developed Bayesian FMMs offer powerful tools for functional data classification.
• Wavelet-based modeling enhances the ability to capture complex functional patterns.
• The robust approach is particularly valuable for datasets with outlying observations.
• These methods represent a significant advancement in functional data analysis for classification.