Overfitting Ratio Calculator
The Gap That Tells You a Model Is Memorizing, Not Learning
A model that scores 98% on training data and 96% on validation data is behaving healthily; one that scores 98% on training data and 70% on validation data has memorized its training set rather than learned generalizable patterns. This calculator quantifies that gap into a single ratio and flags where it lands on a standard risk scale.
The Formula
Gap = Training Accuracy − Validation Accuracy
Overfitting Ratio = (Gap / Training Accuracy) × 100
Overfitting Ratio = (Gap / Training Accuracy) × 100
Worked Example
| Step | Calculation | Result |
|---|---|---|
| Gap | 95% − 82% | 13.00 points |
| Overfitting ratio | (13.00 / 95) × 100 | 13.68% |
Risk Thresholds
| Gap (points) | Assessment |
|---|---|
| 0 – 5 | Low overfitting risk |
| 5 – 10 | Moderate overfitting risk |
| Above 10 | High overfitting risk |
A 13-point gap in the worked example above lands in the "high overfitting risk" band — a signal to consider more regularization, more training data, or a smaller model.
Where This Matters
- Model selection — comparing the overfitting ratio across candidate architectures helps choose one that generalizes rather than the one with the flashiest training accuracy.
- Deciding when to add regularization — a widening gap over successive training runs is a common trigger for adding dropout or weight decay.
- Early stopping decisions — tracking this ratio epoch by epoch shows the point where validation performance stops improving even as training accuracy keeps climbing.
How to Use This Calculator
- Enter the Training Accuracy (%).
- Enter the Validation Accuracy (%).
- Select Calculate to see the point gap, the overfitting ratio, and the risk assessment.
Related Calculations
Add regularization and see its effect with the Dropout Rate Calculator, or dig into the underlying predictions with the Model Accuracy Calculator.