Figure 2.

Comparison of different classification approaches for WHO grade and methylation class. Different approaches for data dimensionality reduction were tested followed by either linear discriminant analysis (LDA) or by an optimized neural network (NN). (A) Receiver operating characteristic (ROC) curve for classification of spectra of the test set using the 30 PC and LDA. (B) ROC curve for classification of spectra of the test set using the 30 PC and NN. (C) Spider plot of the area under the ROC curve (AUC) for different combinations of data dimensionality reduction. The type of classifier (LDA or NN) used and the ground truth (WHO grade or methylation class) is indicated by different lines types as shown in the legend.