Optimisation of window size and feature selection. (a) Effect of window size on mean classification AUC. Mean AUC of 100 repeats of 5-fold stratified cross validation was reported. Two standard errors are shown as error bars. (b) Feature selection using mRMR and incremental feature selection strategy. Number of features to be retained was determined using the mean classification AUC of 100 runs of 5-fold stratified cross validation as the main performance measure. X-axis represents the number of features used in classification. Two standard errors are represented by error bars in the graph. Maximum AUC was found to be 0.95 at 49 features.