Fig 5. Gain in computational and statistical efficiency due to feature pre-selection.

The classification accuracy for the Random Forest algorithm (y-axis; black dots) was calculated over the training data and for different amounts of cross-correlation features (x-axis; grey bars). The choice of the quantity of features to be used as PROTAX-Sound predictors was based on the configuration which showed the highest classification accuracy based on as few features as possible (dashed blue line).