Calculator Inputs
This calculator uses an interpretable XGBoost-style approximation for scenario testing. It helps estimate a boosted margin, class probability, or regression-style score from weighted normalized features.
Example Data Table
| Scenario | Age Score | Income Stability | Account Tenure | Utilization | Recent Delinquencies | Savings Ratio | Estimated Probability | Predicted Output |
|---|---|---|---|---|---|---|---|---|
| Applicant A | 58 | 78 | 36 | 62 | 1 | 28 | 0.3180 | Negative |
| Applicant B | 42 | 88 | 52 | 35 | 0 | 40 | 0.6640 | Positive |
| Applicant C | 71 | 55 | 14 | 84 | 3 | 12 | 0.1190 | Negative |
Formula Used
This page uses a transparent XGBoost-style scoring approximation for fast statistical scenario analysis. It is designed for exploration, not as a substitute for a trained production model.
1. Normalize each feature
scaledi = clamp((valuei − mini) / (maxi − mini), 0, 1)
2. Build feature impact
impacti = scaledi × weighti × gaini
3. Calculate ensemble scale
treeScale = eta × (√trees / 10) × subsample × colsample / (1 + gamma + depth / 10)
4. Penalize model complexity
penalty = lambda × depth / 10 + alpha × mean(|impacti|)
5. Compute final margin
margin = baseScore + treeScale × Σimpacti − penalty
6. Convert to final output
Classification: probability = 1 / (1 + e−margin)
Regression style: predicted value = margin
How to Use This Calculator
- Choose classification when you want a probability and decision label. Choose regression style when you want a direct score.
- Enter ensemble settings such as base score, eta, tree count, depth, gamma, and regularization values.
- For each feature, enter a name, current value, minimum, maximum, weight, and gain multiplier.
- Use positive weights for supportive factors and negative weights for risk-heavy factors.
- Click Predict Now to display the result above the form.
- Review the metrics, contribution table, and Plotly graph. Use the export buttons to save results as CSV or PDF.
FAQs
1. Is this an exact XGBoost model?
No. It is a transparent approximation inspired by boosted additive scoring. It is useful for scenario analysis, education, and rapid comparisons when a trained model is unavailable.
2. When should I use classification mode?
Use classification mode when the target is binary, such as approve or reject, churn or stay, pass or fail. The calculator returns a probability and label.
3. What does the base score represent?
Base score is the starting margin before feature effects and penalties are applied. It acts like the model intercept in a simplified boosting setup.
4. Why do I need minimum and maximum values?
Minimum and maximum values normalize raw inputs onto a comparable scale. That makes weights and gains easier to interpret across different feature units.
5. What do negative feature weights mean?
A negative weight means the feature pushes the prediction downward. Higher values of that feature reduce the margin, probability, or final score.
6. How do gamma, alpha, and lambda affect results?
They reduce aggressive scoring. Gamma moderates ensemble scale, while alpha and lambda create sparsity and smoothness penalties that keep results more conservative.
7. What does the contribution graph show?
The graph shows each feature’s contribution to the final margin after scaling, weighting, gain adjustment, and ensemble shrinkage. Positive bars help; negative bars hurt.
8. Can I use this for real decision automation?
Use it for planning and sensitivity analysis first. Real automation should rely on validated datasets, trained models, bias checks, and domain review.