Macro vs weighted metrics in classification for machine learning

Macro vs weighted metrics in classification for machine learning 

Metric Type	Meaning
Macro	Calculates metric individually per class, then takes the unweighted average. Treats all classes equally, regardless of size.
Weighted	Calculates metric per class, then takes the average weighted by class support (number of true samples per class). Large classes have more influence.

Example Dataset:

Class	True samples (support)	Precision	Recall	F1-score
Cat	50	0.9	0.8	0.85
Dog	40	0.6	0.7	0.65
Tiger	10	0.5	0.6	0.55

Final Result:

Metric	Macro Average	Weighted Average
F1-score	0.683	0.74

When to use and What?

Scenario	Recommended Metric
Class imbalance exists	Weighted
You care equally about all classes	Macro
You want to compare per-class fairness	Macro
You want an overall accuracy-type proxy	Weighted

Coding:

from sklearn.metrics import f1_score

# Example:

f1_macro = f1_score(y_true, y_pred, average='macro')

f1_weighted = f1_score(y_true, y_pred, average='weighted')

Md. Alamgir Hossain

MSc in ICT, IICT, BUET