Figure 3.

Mean-Squared error for 10-fold cross-validation according to the log of lambda on the training lung cancer dataset. (Left) The cross-validation errors and the upper and lower standard deviation along the lambda values of the Least Absolute Shrinkage and Selection Operator (LASSO) regression model are shown. The vertical dotted lines represent the two selected lambdas. The lambda.min value (left line) minimizes the prediction error (MSE), whereas lambda.1se (right line) gives the most regularized model (most simple model within one standard deviation of the optimal model). Values above the plot show the number of variables included in the model. (Right) Confusion matrix depicting the diagnostic potential of the signatures validated on the test dataset (0 = healthy, 1 = tumor).