Home APS Downloads Buy APS Support Register Contact us Knowledge Base of APS Subscribe to the GEPList   Visit GEP

The APS Environment Evaluating the Accuracy of a Model   APS 3.0 offers an extremely complete package of statistical tools which you can use to evaluate the accuracy of your models without leaving the APS modeling environment. To Evaluate the Accuracy of a Model with APS 3.0  In the Run Panel, during a run, check the generation values of Best Fitness and R-square. During a run, both the fitness and R-square of the best model of each generation are shown and plotted in a chart. In the Run Panel, after a run, check the best-of-run values of Fitness and R-square. After a run, both the fitness and R-square of the best model discovered in the run are given both for the training and testing sets. Since the testing set is not used during training, the values of fitness and R-square on the testing set are a good indicator of the generalizing capabilities of your model. In the Results Panel, check the values of additional statistical functions. Depending on the problem category, a varied set of statistical evaluations of your model are shown to you while you are analyzing your new model. These statistics, performed both for the training and testing sets, include: mean squared error, root mean squared error, mean absolute error, relative squared error, root relative squared error, relative absolute error, correlation coefficient, and R-square for all types of problem category; and the classification error, classification accuracy, confusion matrix (true positives, true negatives, false positives, and false negatives), sensitivity, specificity, positive predictive value, and negative predictive value for Classification problems. In the Report Panel, check the values of all the statistical functions performed by APS for your active model. Depending on the problem category, a varied set of statistical evaluations of your model are shown in the Report Panel. These statistics, performed both for the training and testing sets, include: mean squared error, root mean squared error, mean absolute error, relative squared error, root relative squared error, relative absolute error, correlation coefficient, and R-square for all types of problem category; and the classification error, classification accuracy, confusion matrix (true positives, true negatives, false positives, and false negatives), sensitivity, specificity, positive predictive value, and negative predictive value for Classification problems. Home | Contents | Previous | Next