How to use
- Enter TP, FP, TN, and FN from one binary classification result.
- Read balanced accuracy beside recall and specificity, not by itself.
- Compare balanced accuracy with plain accuracy to see whether class imbalance is flattering the headline number.
Wave 6 classification metrics
Equal weight for positives and negatives
Balanced accuracy averages recall and specificity. That makes it a quick way to check whether one class is being treated fairly when the dataset itself is unbalanced.
Inputs
Run a calculation to compare balanced accuracy with recall, specificity, precision, and plain accuracy.
Balanced accuracy is a fairness check across classes
Balanced accuracy is the average of recall and specificity. That means the positive and negative classes each get equal weight, even when the dataset itself is heavily skewed toward one class.
Why compare it with plain accuracy?
If plain accuracy is much higher than balanced accuracy, the model is probably leaning on the dominant class. That is often a sign that minority-class behavior still needs work.
What to inspect next
Balanced accuracy is a good headline, but it is not the whole story. Keep recall, specificity, and the raw confusion-matrix counts visible so stakeholders can see which side of the trade-off is driving the result.
Frequently asked questions
What does balanced accuracy measure?
Balanced accuracy is the average of recall and specificity. It gives equal weight to the positive and negative classes, so it is often more informative than plain accuracy when classes are imbalanced.
Why can plain accuracy be higher than balanced accuracy?
If one class dominates the dataset, a classifier can achieve high plain accuracy mostly by getting the dominant class right. Balanced accuracy reveals whether the minority class is being handled fairly by averaging recall and specificity.
Is balanced accuracy enough on its own?
Not always. Balanced accuracy is a useful headline, but you should still inspect recall, specificity, and the raw confusion-matrix counts to understand the operational trade-off.
Does the share URL include my counts or labels?
No. The share URL stores only lightweight settings such as decimal places. Counts and custom labels stay in your browser.
Related
Comments (optional)
To reduce load, comments are fetched only when needed.