After a Neural Network being trained, it will be evaluated to make sure that it can cope with completely new data-set. Depending on Neural Network application, appropriate evaluation techniques will be used.

Evaluation techniques in pattern recognition application

ANNHUB supports both Confusion Matrix and ROC curve techniques to evaluate the trained Neural Network. As shown in Figure 3.16, the Confusion Matrix for both training, validation and test data-set are presented in great details. Valuable information such as total number of samples in a certain class are correctly identified, and how many samples are miss-classified. This information is summarized in Accuracy, Sensitivity and Specificity table.

Figure 3.16: Confusion Matrix to evaluate trained Neural Network.

ROC curves for training, validation and test data-set are also shown in in Figure 3.17. ROC curves give designers important information regarding to different levels of threshold.

Figure 3.17: ROC curves to evaluate trained Neural Network

Evaluation techniques in function approximation/fitting application

ANNHUB supports Regression curve that present how well the predicted outputs approximate their targets. The Regression curves for training, validation and test data-set are all available.

Figure 3.18: Regression curve to evaluate trained Neural Network

Notes: If the Bayesian Regularization training algorithm is selected, the evaluation information for validation set will not be displayed as that data-set is not required.