Comparison of model performance with regard to the number of used features. The root mean square error and normalized median absolute deviation of the results from the DR7+9 RF experiments are presented. As reference line the performance achieved with the Classic₁₀ features is shown. As it can be seen, from the fourth feature on, the performance of the subsets outperforms the Classic₁₀ features. After the ninth feature, the improvement settles. When adding many more features, the performance will start to degrade.