2.1 HS improves prediction performance of RF (comparison of hs vs mtry vs depth parameters)
4.4. HS Improves Prediction Performance for RF
The results, displayed in Fig 4D, show that HS significantly improves the prediction accuracy of RF across the datasets we considered. Moreover, HS clearly outperforms the two other RF regularization methods (using depth and mtry) in all datasets. This is especially promising because HS is also the fastest and easiest method to implement, as it does not require refitting the RF. Moreover, hsRF tends to achieve its maximum performance with fewer trees than RF without regularization; as a consequence, RF with HS is of-ten able to achieve the same performance with an ensemble that is five times smaller, allowing us to achieve large sav-ings in computational resources.
2.1 HS improves prediction performance of RF (comparison of hs vs mtry vs depth parameters)