Perform a Comprehensive ROC Curve Analysis for Model Evaluation

Act as a statistical learning expert. I need a thorough ROC (Receiver Operating Characteristic) curve analysis for my binary or multi-class classification model. **Project Details:** - Task: [BINARY/MULTI-CLASS] classification - Classes: [LIST_CLASS_NAMES] - Domain: [DOMAIN — e.g., medical diagnosis, fraud detection, churn prediction] - Model(s) evaluated: [LIST_MODELS] - AUC scores obtained: [LIST_AUC_SCORES_PER_MODEL] - Class prevalence: [POSITIVE_CLASS_PERCENTAGE]% positive, [NEGATIVE_CLASS_PERCENTAGE]% negative **Specific Concerns:** - [DESCRIBE_ANY_SPECIFIC_CONCERNS — e.g., model seems overconfident, classes are highly imbalanced] Please provide the following comprehensive analysis: 1. **AUC Score Interpretation:** Interpret each model's AUC score in practical terms. What does an AUC of [AUC_VALUE] actually mean for my use case? Go beyond "probability that a random positive ranks higher than a random negative." 2. **ROC Curve Shape Analysis:** Explain what different ROC curve shapes indicate (hugging the top-left corner, bowing, crossing curves, etc.) and what I should look for in my curves. 3. **Operating Point Selection:** Help me choose an optimal operating point (threshold) on the ROC curve based on my domain requirements. Discuss Youden's J statistic, cost-sensitive selection, and minimum sensitivity/specificity constraints. 4. **Multi-Model Comparison:** If comparing multiple models, explain when AUC alone is insufficient and when ROC curve dominance matters. 5. **Limitations and Pitfalls:** Explain when ROC analysis can be misleading, particularly regarding [CLASS_IMBALANCE_RATIO] class imbalance, and when precision-recall curves should be preferred. 6. **Visualization Code:** Provide Python code using matplotlib and scikit-learn to plot publication-quality ROC curves with confidence intervals using bootstrapping, including proper multi-class handling (one-vs-rest or one-vs-one) if applicable. 7. **Statistical Comparison:** Provide code and methodology for DeLong's test to statistically compare AUC values between two models. Include interpretive commentary that I could use in a technical report or presentation.