Aplicação de técnicas multivariadas para análise dos escores de classificação dos candidatos ao concurso vestibular 2007 da UFSM

Abstract

The present work arises from the necessity of a statistic key which helps the COPERVES in the analysis of the classification methodology of the candidates to the Vestibular. The Discriminant Analysis of Fisher is a well knowed technique to data classification, which in this work will employ the individual scores of the candidates for matter with the aim to value the discrimination rate between candidates through of multivarieds techniques. The capacity of discrimination between the candidates can be understood like an index, which is the value of the discriminant function calculated for each candidate, based on their objective tests and composition marks. According to these marks, were obtained two discriminant functions and three classification functions for three groups (no selecteds, selecteds and successfuls), and a discriminant function and two classification functions for two groups (selecteds and successfuls). The value of the classification function will determine the position of the candidate in one of the three possible situations. In this way, the Discriminant Analysis shows an excellent capacity to discriminate when the groups are samples of medium and large dimensions with reference to the number the respected parameters, and low capacity to classify when the groups are small samples because show a low accuracy in the respected parameters of the discriminant functions, going down the final classification accuracy.