An Analysis on Better Testing than Training Performances on the Iris Dataset

The Iris dataset is a well known dataset containing information on three different types of Iris flowers. A typical and popular method for solving classification problems on datasets such as the Iris set is the support vector machine (SVM). In order to do so the dataset is separated in a set used for training and a set used for testing. The error rate, after training, for the training set should be lower than the error rate on the test set. However, in this paper we show that when solving the classification problem for the Iris dataset with SVMs this is not the case. Therefore, we provide an analysis of the Iris dataset and the classification models in order to find the origin of this interesting observation