Figure 5.

Mean average error (MAE) of Machine Learning (ML) models for the prediction of t-tau. The results are shown for three different subsets. All the models were optimized and evaluated using the PCA transformed subsets, except a Bagging Regressor that was trained using only the features considered by the PCA algorithm as relevant. The models evaluated were: a Bagging Regressor (BR), an Extra Trees Regressor (ETR), a Random Forest Regressor (RFR), a Gradient Boosting Regressor with a least absolute deviation loss (GBR_lad), a least squares loss (GBR_ls) and a Huber loss (GBR_hb), a Ridge Regression (RR), a Lasso Regression (LR), a Gaussian Process with a radial basis function kernel (GP_rbf) and a Matern 3/2 kernel (GP_m32), k-nearest neighbors regression (KNR), Support Vector Regression with a linear (SVR_lin), polynomial (SVR_poly), radial basis function (SVR_rbf), and sigmoid (SVR_sig) kernels.