The fundamental role of density functions in the binary classification problem

In biomedicine, binary classification problems are involved in diagnostic but also, for instance, in personalized medicine. The objective is to use information for correctly allocating subjects in groups. Frequently, this information implies high-dimensional data. An adequate classification rule is...

Full description

Saved in:
Bibliographic Details
Published in:Journal of statistical computation and simulation Vol. 92; no. 13; pp. 2846 - 2861
Main Author: Martínez-Camblor, Pablo
Format: Journal Article
Language:English
Published: Abingdon Taylor & Francis 02.09.2022
Taylor & Francis Ltd
Subjects:
Asymptotic properties
Binary classification problem
Classification
Density
Distribution functions
kernel density estimator
machine learning
Probability distribution functions
robust estimator
Support vector machines
ISSN:0094-9655, 1563-5163
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In biomedicine, binary classification problems are involved in diagnostic but also, for instance, in personalized medicine. The objective is to use information for correctly allocating subjects in groups. Frequently, this information implies high-dimensional data. An adequate classification rule is a trade-off between the sensitivity and the specificity. The ROC curve helps to understand, evaluate and compare the accuracy of classification processes. We propose a procedure for estimating the optimal classification rules based on a penalized estimator of the underlying probability distribution functions. We study its asymptotic properties. Through Monte Carlo simulations, we compare our proposal with a support vector machine-based ROC curve. We illustrate its practical use in a real-world problem. Results suggest that, despite some techniques promise to improve the results provided by traditional methods, in the binary classification problem, the limit is the actual relationship among the density functions.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0094-9655
1563-5163
DOI:10.1080/00949655.2022.2051026