Caracterización por árboles de decisión para datos categóricos: diagnóstico de virus respiratorios en aves.

Abstract

This document aimed to identify the need to implement new diagnostic tests for undetermined viral diseases in birds, in the case of a molecular biology veterinary diagnostic laboratory in Bogotá, Colombia. Thus, through the purification of the database and its subsequent analysis by decision trees CHAID, Random Forest and Logistic Regression. The CHAID tree methodology was applied to identify the most important variables in relation to the response variable. The results of this indicate that the variables with the greatest impact on the response variable are the variables of Department, Technique and Test requested, which coincides with the results obtained by Random Forest and by the Logistic Regression model. Regarding the latter, a binary model of the response variable (Result) was proposed with two options: positive or negative and with 6 factors as independent variables. The three methodologies applied account for the imbalance of the data, with a high specificity due to the negative level of the Result factor, which responds to the objective set, since the need to implement new diagnostic tests for undetermined viral diseases in birds is identified.