Fig. 4.
Visualisations of model evaluation. Predicted scores (train and test) split into the respective binary classification, visualised in three different ways. a, b Violin plots; c, d probability distribution function (pdf) plots. Red, healthy controls (control); blue, gastric cancer (case). e, f ROC curves with 95% CIs derived from 100 iterations of bootstrap resampling. Green line predicted scores for training set; green 95% CIs, IB predictions; yellow line, prediction scores for test set; yellow 95% CIs, OOB predictions. PLS-DA AUCTrain = 0.97, AUCTest = 0.89, AUCIB = 0.92–0.99, AUCOOB = 0.72–0.98. ANN AUCTrain = 1.00, AUCTest = 0.90, AUCIB = 0.95–0.99, AUCOOB = 0.77–1.00