Support vector machines, Naïve Bayes classifier and genetic algorithms for the prediction of psychosocial risks in teachers of Colombian public schools.

Abstract

This paper presents a new methodology based on the application of Support Vector Machine algorithms, Naïve Bayes and Genetic Algorithms in diagnostics of psychosocial evaluations for the identification and prediction of the psychosocial risk level of public-school teachers in Colombia. A comparative study of the model of machine learning for prediction was carried out: Support Vector Machines (SVM) and Naïve Bayes, in two stages, first with all the variables and second, reducing the dimensionality of the database applying genetic algorithms, The best forty variables with the best efficiency in prediction accuracy were selected. The database used consisted of 3000 epidemiological records, which corresponded to teachers from public schools in the metropolitan area of a Colombian city. The use of SVM easily detected variables of physiological type and the best prediction performance was obtained with accuracy of 96.3%.