Fig. 3From: Development of models for predicting Torsade de Pointes cardiac arrhythmias using perceptron neural networksThe ROC curves (cumulative) of SDAR for the training, test and validation (38 blind hold-out test drugs) sets. The majority of these models (overlapped blue lines) indicated good predictive accuracy. The lines closer to the diagonal obviously come from the predictions in the internal and external training sets, less accurate than those of the training setsBack to article page