There seems to be different opinions about using one-hot encoded categorical features with implementations that don't natively support them. Try CatBoost or H20 Random Forrest that support categorical data by design. Also, investigate one-hot encoding not being recommended for features with high cardinality, something to do with creating very sparse features.
For reference :
https://roamanalytics.com/2016/10/28/are-categorical-variables-getting-lost-in-your-random-forests/
https://medium.com/data-design/visiting-categorical-features-and-encoding-in-decision-trees-53400fa65931
https://www.kaggle.com/c/avito-demand-prediction/discussion/57094
https://www.kaggle.com/c/zillow-prize-1/discussion/38793
https://www.kaggle.com/c/bnp-paribas-cardif-claims-management/discussion/19851 \
Resolution : Looking at the Feature Importance for the models.
- Random Forrest - just seems to put all importance on the size feature and the categorical features are drowned out.