Fig. 4

Prediction performance plots for overall survival. a Prediction performance plot for overall survival of the combined oral tongue cohort of the mean value derived from the machine learning method RFE-SVM using a range of parameters (lambda range 0.01 with a poly kernel). Error bars represent the range of values obtained across the three-fold cross validation performed. The prediction performance ranged between 0.5 and 0.58 indicating a poor performance. b Representative prediction performance plot for overall survival of the combined oral tongue cohort using CFS. The training set demonstrates a decrease in performance with an increasing number of probes utilised, and the test model derived from the training set shows poor performance. Error bars represent the range of values across the three-fold cross validation performed. For both plots, a performance of 1.0 represents a 100 % chance of being able to classify groups according to the nominated variable, according to the number of probes interrogated. A prediction performance of 0.5 represents a capability of no better than chance