I don't know if anyone is as confused as I am about how the evaluation metric for this competition is combined. I understand that it is a multimetric method, but I am a bit confused about the direction that indicates a better model.

At the info tab, it is said that the score is a weighted average of MAE and RMSE. If this is the case, I think a lower score indicates a better model, since both metrics are minimised (errors). However, on the leaderboard, it is not; rather, a higher score indicates a better model.

Let's assume in a hypothetical case that a model has MAE and RMSE values of 5 and 7, respectively and another, 2 and 5, respectively. A weighted average (equal weighting) of the two models would be 6 and 3.5, respectively. In this case, the second model is better than the first, but in the case of the leaderboard, the first model is better.

I don't know, but maybe I may be wrong with how the scores are combined. Probably, @Zindi may show us a formula on how both metrics are averaged.

Peace!