Results
Step-by-step calculation
Per-class metrics
| Class | Support | Correct | Accuracy | Recall |
|---|
Confusion matrix heatmap
Training run comparison
| Run | Training | Validation | Gap | Assessment |
|---|
Merged report data
Formula used
How to use
Select a calculation mode matching your available data. Enter valid nonnegative values in each required field. Click calculate to review metrics, charts, and interpretation clearly.
Use validation accuracy to inspect possible overfitting. Add run data to compare several model versions. Export the completed report using available action buttons easily.
Example data
| Input | Value | Meaning |
|---|---|---|
| Correct predictions | 850 | Samples classified correctly. |
| Total samples | 1000 | All training observations. |
| Training accuracy | 85% | Eight hundred fifty predictions were correct. |
| Error rate | 15% | One hundred fifty predictions were incorrect. |
Calculation history
| Time | Mode | Accuracy | Samples | Validation | Gap |
|---|---|---|---|---|---|
| No saved calculations. | |||||
Frequently asked questions
What is training accuracy?
Training accuracy measures correct predictions on training data. It summarizes model fit using a simple proportion. High accuracy alone cannot confirm strong generalization performance reliably.
How is training accuracy calculated?
Divide correct predictions by total training samples. Multiply the decimal result by one hundred. The calculator also displays the complementary error rate clearly.
What is balanced accuracy?
Balanced accuracy averages recall across all classes. It reduces dominance from very frequent classes. This makes comparisons fairer for imbalanced classification datasets overall.
Why compare training and validation accuracy?
The comparison reveals possible generalization problems. A large positive gap can indicate overfitting. Similar values usually suggest more consistent model behavior overall.
What is top-k accuracy?
Top-k accuracy checks several ranked predictions per sample. A result counts when truth appears among them. It is common for many-class prediction problems in practice.
Can accuracy mislead on imbalanced data?
Yes, majority classes can dominate standard accuracy. Review balanced accuracy and per-class recall together. Precision and F1 scores may provide better context overall.
What does the confidence interval show?
It estimates uncertainty around observed accuracy. Larger sample sizes usually narrow the interval. The calculator uses a Wilson proportion interval method internally.
What is weighted accuracy?
Weighted accuracy gives selected samples or classes greater influence. It supports unequal importance or sampling corrections. Weights should reflect a defensible evaluation objective in practice.
Does high training accuracy prove success?
No, memorization can produce very high training accuracy. Always evaluate untouched validation or test data. Review calibration, robustness, and task-specific costs as well carefully.