model-comparison

To compare models fairly, this Skill runs XGBoost, LR+WoE, and DNN side-by-side on a shared train/val/OOT data split, aggregating OOT AUC, KS, BCR, and Brier scores to generate an objective multi-model evaluation report.

A Pareto frontier candidate set identifies algorithms on the Pareto front without subjective weighting or gatekeeping, leaving the final model selection to LLM or business reasoning based on objective metrics.

You measure model stability by calculating KS gap and PSI drift alongside OOT metrics, and performing DeLong AUC significance tests to evaluate binary scoring model stability across time periods.

Yes, you can run multi-algorithm comparisons for general, scorecard, fraud, and stability-first scenarios by applying consistent data slicing and scenario context to evaluate which model is better.

No, you do not need tuned hyperparameters to run the comparison, but the Skill optionally loads them to refine the XGBoost, DNN, and LR+WoE models during the multi-dimensional evaluation process.