Figure 1
From: Multiple Kernel Learning Model for Relating Structural and Functional Connectivity in the Brain
Comparison of Model Performance on Individual Test subjects. (a) Pearson correlation between empirical and predicted FCs of all the test subjects by multiple kernel learning (MKL) model and performance comparison with the predictions by the other two models. While MKL model has superior performance compared to that of dynamic mean field (DMF) and single diffusion kernel (SDK), DMF model performs slightly better than the SDK model. (b) Results of leave-one-out cross-validation on the test subjects also yield similar comparative performance. Note that the subject indices are kept identical between sub-figures (a) and (b). This plot suggests that MKL model can handle an increase in the number of training subjects without necessarily any over-fitting. (c) Box-plots of Pearson correlation measure on 9 randomly chosen validation subjects for each of the 5 folds for the MKL model. Points lying outside the quartiles are the suspected outliers. Compactness of boxes suggests inter-subject consistency of model’s performance. Further, these 5-fold cross-validation results suggest that MKL model performs consistently well on unseen subjects.
