Closed pskoland closed 7 months ago
Based on the winning approach from a similar competition (https://www.kaggle.com/competitions/ams-2014-solar-energy-prediction-contest/discussion/6321), try training separate models for each location, and then averaging the results. This should (according to the winning approach) optimize MAE, even without much feature engineering.
Based on the winning approach from a similar competition (https://www.kaggle.com/competitions/ams-2014-solar-energy-prediction-contest/discussion/6321), try training separate models for each location, and then averaging the results. This should (according to the winning approach) optimize MAE, even without much feature engineering.