• Because of the log you don’t penalize huge differences between predicted and the actual values, especially when both are large numbers.
• What you care the most about when you use RMSLE is the scale of your predictions with respect to the scale of the ground truth.
• ML algorithms can better optimize if you apply a logarithmic transformation to the target before fitting it (and then reverse the effect by using the exponential function). Siehe dazu auch den Abschnitt TransformedTargetRegressor in RMSE.
• Most used Metric for regression on Kaggle
  • Recruit Restaurant Visitor Forecasting
  • Sberbank Russian Housing Market
  • Mercari Price Suggestion Challenge
  • Santander Value Prediction Challenge
  • How much energy will a building consume?