Table 2 Confusion matrix for the prediction of heart disease for the (first) neural network using the four CK metabolic parameters, [PCr], [ATP], kf, and CK flux). The numbers at the top of the boxes are the number of cases classified correctly (green boxes), or misclassified (orange boxes): their sum is the total classified (166 = 178–12 HCM patients who had HF and were omitted from the training set). The percentages of the total are listed below the numbers (eg, for HF, 87/166 = 52% of the sample). At bottom, the % correct (true positive rate, green bold font) and incorrect (false negative rate, red bold font) are summarized for each output disease class (HF, HCM, MI and Control). For example, in the first column 87 HF cases were correctly classified. An additional 10 HF cases were incorrectly assigned as non-HF (1 as MI and 9 as Controls). Thus 87/97 = 90% are correct and 10% are incorrect. The percentages in the right-hand column (positive predictive value, green font; false discovery rate, red font) include subjects from all classes. For example, a total of 102 cases were classified by the network as HF. This comprises 87 true HF cases, plus 1 HCM, 5 MI and 9 Controls all misclassified as HF. Thus, 87/102 = 85% are correct. Summing the diagonal, 140/166 = 84% of cases are correctly classified overall
