A comparative study of multiple discriminant analysis and multinomial logistic regression applied to students' performance

Abstract

This study compared the performance of two of the most recommended statistical techniques in classification. The study sought to determine the classification accuracy of Multiple Discriminant Analysis (MDA) and Multinomial Logistic regression (MLR) by classifying the students based on their performances on the modules offered in Statistics Department at the North West University. The data used comprised of the performance of the third year students who majored in Statistics from 2013 and 2016 academic years. The preliminary analysis were performed to evaluate the descriptive statistics and test for assumptions. In order to achieve the MDA results, the Wilk's lambda was used to test the significance of the model. The canonical discriminant function coefficients were utilised to construct a discriminant model while the classification table was used to generate the overall classification of the students' performances. The model fitting table MLR showed that the model was useful. The deviance and the Pearson's statistic were used to indicate if the model fits the data or not and the parameter estimates were used to obtain the fitted model for MLR. The results from the classifications showed that the MDA was able to classify 58.3% and the MLR was able to classify 56.7%. These results showed that the techniques do not differ much in classification. Therefore both the techniques can be used for future studies in education.