A comparison between some discriminative and generative classifiers (Logistic Regression, Support Vector Machines, Neural Networks, Naive Bayes and Bayesian Networks).

The purpose of this thesis is to compare different classification methods, on the basis of the results for accuracy, precision and recall. The methods used are Logistic Regression (LR), Support Vector Machines (SVM), Neural Networks (NN), Naive Bayes(NB) and a full Bayesian network(BN). Each section describes one of the methods, including the main idea of the methods used, the explanation of each one, the intuition underpinning each method, and their application to simple data sets. The data used in this thesis comprises 3 different sets used previously when learning the Logistic Regression model and the Support vector Machines one, then applied also to the Bayes counterparts, also to the Neural Networks model. The results show that the Bayesian methods are well suited to the classification task they are as good as their counterparts, some times better. While the Support Vectors Machine and Neural Networks are still the best all around, the Bayesian approach can have comparable performance, and, makes a good approximate to the traditional method's power. The results were Logistic Regression has the lowest performance of the methods for classification, then Naive Bayes, next Bayesian networks, finally Support Vector Machines and Neural Networks are the best.