Using Logistic Regression as a Classifier in Modeling with Normal Mixtures رسالة ماجستير

Abstract

The classification of observations plays an important role in statistics and all other fields. In this thesis, we studied Logistic Regression (LR) as a method of classification and compare its performance with the performance of Linear Discriminant Analysis (LDA), Gaussian Mixture Model (GMM), and Neural Networks (NN). Performance is compared by the Misclassification Table and Error Rate for each method. Furthermore, the effect of sample size and presence of correlation were studied. In general, the results showed that when the linear discriminant analysis assumptions are met, the performance of the linear discriminant analysis method is best. If the conditions are not met, the logistic regression method outperforms the other classification methods.