Evaluation of ANN Classifiers During Supervised Training with ROC Analysis and Cross Validation
Miguel Antônio Sovierzoski, Fernanda Isabel Marques Argoud, Fernando Mendes de Azevedo · 2008
The evaluation of an Artificial Neural Network is not a part of the training phase and it is not a trivial process. It represents an exhaustive test process with a computational effort greater than the ANN training. Monitoring the error during the training phase can provide an indicator of the convergence of the algorithm. This study presents some analysis tools integrated to the supervised training of the ANN MLP Classifier. The objective of this study is to provide a quantitative evaluation of the learning and generalization of the knowledge during the ANN supervised training. The Cross Validation and the ROC Analysis procedures were used together with the standard back-propagation ANN MLP training algorithm. The procedure is described and the results of the ANN classifier for epilepsy events in EEG data are presented.