Figure 11
Evaluation of multi-variate models using the Monte Carlo cross validation (M = 50) for IPI response prediction. (A) Comparison of methods regarding their aggregated stratification performance: (1) uni-variate stratification with 10-times repeated 8-fold cross validation as reference, (2) tree (d = 1, smin = 5, lmin = 3), (3) tree (d = 2, smin = 5, lmin = 3), (4) tree (d = 2, smin = 5, lmin = 5), (5) tree (d = 2, smin = 5, lmin = 5), information gain measure: cross entropy, (6) tree (d = 2, smin = 8, lmin = 6), (7) tree (d = 3, smin = 5, lmin = 3), (8) SVM, 5 features, hinge, (9) SVM, 3 features, hinge, (10) SVM, 5 features, hinge sqr, (11) SVM, 7 features, hinge, (12) perceptron, 5 features, (13) logistic regression, 5 feat, (14) bi-variate stratifier with 10-times repeated 8-fold cross validation (all functions), (15) bi-variate stratifier with 10-times repeated 8-fold cross validation (only AND). (B) Aggregated Kaplan-Meier plot (with respect to OS and stratification according to predicted IPI response) for the method (4) tree (d = 2, smin = 5, lmin = 5) yielding the highest accuracy. Blue: predicted response, gray: predicted non-response.
