pone.0278388.s003.tif (1.92 MB)

Random forest model performs better than support vector machine algorithms and when it primarily uses spontaneous photopic ERG of 60-s duration in humans.

figure

posted on 2023-01-12, 18:41 authored by Ramsés Noguez Imm, Julio Muñoz-Benitez, Diego Medina, Everardo Barcenas, Guillermo Molero-Castillo, Pamela Reyes-Ortega, Jorge Armando Hughes-Cano, Leticia Medrano-Gracia, Manuel Miranda-Anaya, Gerardo Rojas-Piloni, Hugo Quiroz-Mercado, Luis Fernando Hernández-Zimbrón, Elisa Denisse Fajardo-Cruz, Ezequiel Ferreyra-Severo, Renata García-Franco, Juan Fernando Rubio Mijangos, Ellery López-Star, Marlon García-Roa, Van Charles Lansingh, Stéphanie C. Thébault

A, ROC curves for both linear and radial svm algorithms. B, Performance parameters for the random forest model using power spectra from photopic or mesopic ERGs of 10, 30 or 60 s. C, ROC curves for the random forest model using power spectra from photopic, mesopic or combined photopic and mesopic ERGs of 60 s. D, Corresponding performance parameters. All data correspond to binary classification between control and disease cases. Controls are constituted by metabolically healthy subjects (n = 62) and the disease group by patients with overweight (n = 41), obesity (n = 16), metabolic syndrome (n = 55), and diabetes with no DR (n = 63).

(TIF)