MODELOS PREDITIVOS DA EVASÃO DE ESTUDANTES

COMPARATIVO ENTRE REGRESSÃO LOGÍSTICA, REDES NEURAIS E ALGORITMOS GENÉTICOS

Abstract

This study aims to perform a technical and comparative analysis of the accuracy of predictive models for student dropout, exploring three distinct approaches: genetic algorithms, neural networks, and logistic regression. An applied and exploratory research was conducted, involving a literature review to validate the relevance of these algorithms in educational contexts and to assess the influencing variables of dropout in higher education. The literature review revealed a significant presence of these algorithms in predictive dropout models and confirmed the pertinence of the variables used. Moreover, the study detailed how the three algorithms were operationalized and presented a comparative performance analysis of the models using data from 447 students of Business Administration and Engineering from a private institution, covering 11 influencing variables of dropout. The results indicate an accuracy of 90% for the model developed with logistic regression, 84.4% for the model developed with neural networks, and 76% for the model developed with a genetic algorithm, highlighting a greater efficiency of the logistic model in predicting student dropout in the sample of this research.