Analyze every class before trusting summary metrics. Find weak labels, skew, and missed positives quickly. Use exports and charts for cleaner model review workflows.
Use one card per class. Large screens show three cards, medium screens show two, and small screens show one.
The chart compares per-class precision, recall, and F1. Submit the form to refresh values.
| Class | TP | FP | FN | Support | Precision | Recall | F1 Score |
|---|---|---|---|---|---|---|---|
| Cat | 42 | 6 | 8 | 50 | 0.8750 | 0.8400 | 0.8571 |
| Dog | 37 | 9 | 11 | 48 | 0.8043 | 0.7708 | 0.7872 |
| Bird | 29 | 5 | 7 | 36 | 0.8529 | 0.8056 | 0.8286 |
| Fish | 24 | 4 | 6 | 30 | 0.8571 | 0.8000 | 0.8276 |
| Macro Average F1 | 0.8251 | ||||||
Macro averaging gives every class equal weight, even when support differs. This helps expose weak performance on rare classes that accuracy may hide.
It averages the F1 score of every class equally. Large classes do not dominate the result, so weak minority-class performance remains visible.
Use macro F1 when class imbalance matters or when every class deserves equal attention. Accuracy can look strong even if rare classes perform badly.
Macro F1 treats every class equally. Weighted F1 multiplies each class F1 by support, so common classes influence the overall score more.
Micro F1 aggregates totals across classes first. Strong performance on common classes can raise micro F1, while macro F1 still punishes weak classes equally.
You need true positives, false positives, and false negatives. The calculator derives support, precision, recall, and F1 automatically from those values.
The calculator applies your chosen zero-division policy. You can return 0 or 1 when TP+FP, TP+FN, or precision+recall equals zero.
Yes, as long as your TP, FP, and FN values are valid for each class. The macro F1 formula is identical once those class-level statistics exist.
Support equals TP + FN. It helps you judge how much evidence backs each class metric and how imbalance may affect interpretation.
Important Note: All the Calculators listed in this site are for educational purpose only and we do not guarentee the accuracy of results. Please do consult with other sources as well.