Kaggle forecasting competitions: An overlooked learning opportunity

Competitions play an invaluable role in the field of forecasting, as exemplified through recent M4 competition. The competition received attention from both academics and practitioners sparked discussions around representativeness data for business forecasting. Several competitions featuring real-life forecasting tasks on Kaggle platform has, however, been largely ignored by academic community. We believe learnings these have much to offer community provide a review results six competitions. find that most datasets are characterized higher intermittence entropy than M-competitions global ensemble models tend outperform local single models. Furthermore, we strong performance gradient boosted decision trees, increasing success neural networks variety techniques adapting machine learning task.