Module 30 final model calibration and decision curve analysis completed.

Final model:
- F_fisherz_elasticnet100_lgbm.
- Independent test sample size: N = 156, ASD = 71, TC = 85.

Discrimination:
- AUC = 0.791
- Balanced accuracy = 0.744
- Sensitivity = 0.746
- Specificity = 0.741

Calibration:
- Brier score = 0.195
- Expected calibration error, 10 bins = 0.118
- Maximum calibration error, 10 bins = 0.268
- Expected calibration error, 5 bins = 0.108
- Maximum calibration error, 5 bins = 0.149

Decision curve analysis:
- The model showed greater net benefit than both treat-all and treat-none strategies across the following threshold probability interval(s):
  0.06–0.06; 0.17–0.77; 0.99–0.99
- Mean net benefit between threshold probabilities 0.20 and 0.80:
  model = 0.1761
  treat-all = -0.2664
  treat-none = 0.0000

Interpretation:
- Calibration and decision-curve results are based on the final upgraded model and should replace previous SVM-based calibration/DCA text.
- These analyses support that the final model contains decision-relevant information in the independent test set, but the results should still be interpreted as exploratory and not as evidence for standalone clinical diagnostic use.