Fig. 6

Binary-group classification results of multimodal models after applying different grouping strategies. RF, SVM, and DNN AUROCs were slightly higher in (N + PPG, G) groups (RF: 99.6%, SVM: 99.7%, DNN: 99.2%) and in (N, PPG + G) groups (RF: 99.6%, SVM: 99.6%, DNN: 99.0%) with tenfold cross validation. A similar result was shown in the test set in (N + PPG, G) groups (AUROCs: RF: 93.8%, SVM: 94.4%, DNN: 95.4%), and in (N, PPG + G) (AUROCs: RF: 93.8%, SVM: 93.3%, DNN: 94.5%). There were decreases in the AUROCs for the test dataset with both methods, but the best AUROC of the three models still exceeds 90%. RF, random forest; Ada, adaptive boosting; SVM, support vector machine; LogReg, logistic regression; NB, Naïve Bayes; KNN, k-nearest neighbor; CART, classification and regression tree; C4.5, C4.5 decision tree; DNN, dense neural network; AUROC, area under receiver operating characteristic curve