Act as a Senior Data Scientist. I am working on a fraud detection dataset where the positive class (fraud) represents only 0.1% of the data. I cannot collect more data. Propose a comprehensive modeling pipeline that includes: 1) Data resampling techniques (SMOTE, ADASYN, etc.) and their trade-offs, 2) Algorithm selection focusing on anomaly detection vs classification, 3) Cost-sensitive learning approaches, and 4) Evaluation metrics that are more informative than Accuracy or ROC-AUC (such as Precision-Recall AUC). Provide Python code snippets using Scikit-Learn and Imbalanced-Learn to demonstrate the pipeline.