Beyond Supervised Learning for Pervasive Healthcare.

Area of Science:

• Computational health informatics
• Machine learning in medicine
• Sensing technologies for healthcare

Background:

• Healthcare data scarcity, quality, and heterogeneity limit supervised machine learning.
• Supervised learning relies on statistical fitting, which is insufficient for complex healthcare data.
• Pervasive healthcare applications require robust and generalizable computational models.

Purpose of the Study:

• Identify challenges in applying machine learning to pervasive healthcare.
• Review learning strategies that overcome limitations of fully supervised learning.
• Propose future directions for data-efficient and trustworthy healthcare AI.

Main Methods:

• Survey of current trends beyond fully supervised learning.
• Analysis of seven key learning strategies to enhance generalization.
• Identification of emerging research directions in healthcare informatics.

Main Results:

• Limitations of empirical risk minimization in supervised learning are highlighted.
• Seven distinct learning strategies are summarized to improve model performance.
• Key challenges in healthcare data are addressed by advanced methods.

Conclusions:

• Advanced learning strategies are crucial for effective machine learning in pervasive healthcare.
• Future research should focus on data efficiency, scalability, and trustworthiness.
• Leveraging multi-modality and multi-source sensing data is essential for future healthcare AI.