Comparison of Classifiers Based on Neural Networks and Support Vector Machines
Pilar Pérez Conde, Irene Carrillo · 2017
This paper presents the compared performance machine learning algorithms specifically classifiers based on neural networks and support vector machines. This comparison was realized with a different dataset of PROMISE Software Engineering Repository (TunedIT) and Weka (Waikato Environment for Knowledge Analysis) software. The main objective is to compared the performance of Bayes Networks, the Radial Base Function (RBF networks), Multilayer perceptron and Support Vector Machines in the classification task using different dataset to determine if the dataset size and the number of attributes or classes influence the performance of the task. The metrics used to measure performance were Accuracy as principal, F-measure, precision, Kappa statistics and ROC curve. The experimental result shows the neural networks as the first best algorithm for classification task with the different dataset achieving and the and the second is the support vector machines, for three datasets, the values for both are 95.8% of accuracy, and 0.84 and 0.85 of Kappa statistics respectively.