Same as squared MSE:
Anders als MSE ist RMSE nicht ganz so streng mit large prediction errors, da hier nochmal die Wurzel gezogen wird.
<aside> 💡 Consequently, depending on the problem, you can get a better fit with an algorithm using MSE as an objective function by first applying the square root to your target (if possible, because it requires positive values), then squaring the results. Functions such as the TransformedTargetRegressor in Scikit-learn help you to appropriately transform your regression target in order to get better-fitting results with respect to your evaluation metric. Siehe dazu den passenden Abschnitt in RMSLE!
</aside>
RMSE on Kaggle: