Two unequal trays of colored classification tokens beside a balanced scale. Original AI-generated artwork produced for GreenCalc; not a measured diagram or empirical data display.
Two averages with different weights
Ordinary accuracy divides all correct predictions by all observations. Each observation receives equal weight, so a large class contributes more to the result. Balanced accuracy instead averages sensitivity and specificity in a binary task. Each actual class contributes half of that score, regardless of how many observations it contains. Neither definition changes the original predictions.
A numerical comparison
Take TP = 8, FN = 2, FP = 12 and TN = 978. Sensitivity is 80%; specificity is 978/990, about 98.7879%. Balanced accuracy is their average, about 89.3939%. Ordinary accuracy is 98.6%. The gap is not a calculation error: the ordinary score gives 990 negative observations their full numerical weight, while the balanced score gives the positive and negative classes equal influence.
| Metric | Calculation | Result |
|---|---|---|
| Accuracy | 986 / 1,000 | 98.6% |
| Balanced accuracy | (0.8 + 978/990) / 2 | 89.3939% |
What balancing does not solve
A balanced score does not establish the cost of false positives, the precision among alerts, or performance for subgroups. It also cannot repair inaccurate reference labels or a dataset that excludes difficult cases. A class with no actual observations has an undefined recall; this tool leaves balanced accuracy undefined in that situation rather than inventing a zero for the missing component.
Report a pair of scores with their denominators
Use the Accuracy Calculator to calculate both scores from the same four counts. State which one informed your comparison and why its weighting suits the task. Keep positive and negative sample sizes in the report, especially when the positive class is small. If the evaluation mixture differs from deployment, examine class-specific rates before transporting a headline score. For a concrete baseline example, read why 99% accuracy can lose to a simple baseline. For precision, F1 and MCC, continue to the full confusion-matrix report.
Check the model behind your result
Use the Accuracy Calculator for the supported calculation and review its stated assumptions before interpreting the output. Find related tools in the Statistics category, or browse the GreenCalc homepage for other calculation topics.
By GreenCalc Editorial Team · Updated 2026-10-26