Classification With the Matrix-Variate-t Distribution

• NCJ Number: 303229
• Journal: Journal of Computational and Graphical Statistics Volume: 29 Issue: 3 Dated: 2020 Pages: 668-674
• Author(s): Geoffrey Z. Thompson; Ranjan Maitra; William Q. Meeker; Ashraf F. Bastawros
• Date Published: January 2020
• Grant Number(s): 2015-DN-BX-K056; 2018-R2-CX-0034
• Agencies: NIJ-Sponsored
• Publication Type: Research (Applied/Empirical)
• Award: Quantitative Tools for Examination of Microscopic Fracture Surface Topology and Degradation for Physical Match Analysis of Evidence; Physics and Statistical Models for Physical Match Analysis Utilizing 3D Microscopy of Fracture Surfaces

Annotation

This article develops an expectation-maximization algorithm for discriminant analysis and classification with matrix-variate t-distributions. 

Abstract

Matrix-variate distributions can intuitively model the dependence structure of matrix-valued observations that arise in applications with multivariate time series, spatio-temporal, or repeated measures. The methodology presented in the current article shows promise on simulated datasets or when applied to the forensic matching of fractured surfaces or to the classification of functional magnetic resonance, satellite, or hand gestures images. (publisher abstract modified)