UK Accidents 10 years history with many variables

Dataset from Kaggle.

The original datasets were large, with 1.6 million, 3M and 2.2 million instances, with 32, 22 and 15 features per section respectively. Then merge Vehicles with Accidents by “Accident_Index” and then merge it with Casualties by (“Accident_Index”, “Vehicle_Reference”). After merging, the final dataset consists of 3.3M rows and 66 columns. “Accident_Severity” is target which is what we need to predict.

Several models were created including Random Forest, Decision tree with bagging, XGBoost, Naive Bayes.