Advanced Machine Learning Feature Engineering for Scientific Datasets

Develop a sophisticated feature engineering pipeline specifically tailored for scientific research datasets with high-dimensional, sparse, and noisy characteristics. Create a solution that can: 1) Implement domain-adaptive feature selection algorithms, 2) Handle non-linear transformation techniques, 3) Generate interpretable feature importance reports, and 4) Support both supervised and unsupervised learning paradigms. Include robust techniques for managing small sample sizes typical in specialized research domains.