Bridging Expert Knowledge and Automated Feature Engineering via Self-Evolution
Researchers have developed FEST (Feature Engineering with Self-evolving Trees), a novel method for automated feature engineering that bridges expert knowledge with machine learning. FEST is designed to create interpretable and discriminative features from unstructured data like text and images, aligning with domain-specific expert criteria. In evaluations across brand compliance, clinical care, and content moderation tasks, FEST outperformed existing methods, demonstrating significant accuracy gains and achieving high semantic alignment with expert-designed features. AI
IMPACT This method could enable more reliable and interpretable AI deployments in high-stakes domains by grounding feature engineering in expert knowledge.