# Which method should be considered to evaluate the imbalanced multi-class classification?

I am working on multiclass-imbalanced data. My dependent variable is highly skewed.

           Injury

2(No Injury)         208753
1(Medium Injury)      22318
0(severe Injury)       3394


I have used random forest algorithm with parameter "class_weight='balanced' " to manage the class 2 imbalance.

I get the below results when I use average='micro'.

 [[   34   107   688]
[  148   778  4592]
[  905  4635 46730]]
Accuracy Score: 0.8110616374089428
precision score: 0.8110616374089428
Recall score: 0.8110616374089428
AUC Score: 0.8582962280567071
F1 score: 0.8110616374089428
Kappa Score: 0.05522284663052324


For the average = 'macro', the results are below.

[[   31   125   684]
[  157   838  4559]
[  890  4694 46639]]
Accuracy Score: 0.8104816009007626
precision score: 0.3586119227436326
Recall score: 0.3602869806251181
AUC Score: 0.5253225798824679
F1 score: 0.3592735337079687
Kappa Score: 0.06376296115668922


So, which results should I consider to evaluate the model? If I have to consider the macro, then my model performance is really bad. Please suggest if there are any methods to improve the precision, recall and AUC score?

If I consider micro results, my precision, recall, f1 score is same. How can I justify this in the project?

I am interested in medium and severe injuries. But individual class precisions 0.0312(severe), 0.1409(medium) for macro are very low. The overall precision score is also very low 0.35. Is it possible to increase these scores? Or Is it fine to consider these low values as my final results for the project? Any suggestions are greatly appreciated. NOTE: I have tried SMOTE oversampling and ensemble cross-validation with different algorithms but I ended up having precision and recall scores less than 50.

Any help would be appreciated.

Thank you.

• If you think that there is a problem to your model by the imbalanceness of the classes, have you considered using a resampling methods such SMOTE for oversappling? – JoPapou13 Nov 7 '18 at 19:45
• Yes, I have used smote method to oversample and then computed the above results. – Bhaskar Sharma Nov 7 '18 at 20:00

• There's also weighted mode for precision, recall and f1. – Piotr Rarus - Reinstate Monica Dec 3 '19 at 12:16
For imbalanced datasets you can employ F1 score. It considers both rare and common classes. You can take a look at this article if you are not familiar with that.