Machine Learning Is Not Just for Prediction: Its Role as an Exploratory Analytical Tool in Medicine

Area of Science:

• Medical research
• Data science
• Biostatistics

Background:

• Machine learning (ML) in medicine is often limited to predictive modeling.
• A broader application of ML involves exploring data to uncover novel insights.
• Conventional statistical methods face limitations with complex, high-dimensional medical data.

Purpose of the Study:

• To introduce a conceptual framework distinguishing predictive ML from exploratory ML.
• To highlight the utility of exploratory ML in medical research where traditional methods are insufficient.
• To propose revised evaluation criteria for exploratory ML studies.

Main Methods:

• Conceptual review and distinction between predictive and exploratory ML.
• Discussion of unsupervised learning, interpretable supervised learning, and network analysis as exploratory ML methods.
• Analysis of the differences between exploratory ML and hypothesis-driven statistics.

Main Results:

• Exploratory ML identifies data structures to generate hypotheses, unlike predictive ML which focuses on accuracy.
• Exploratory ML is valuable for high-dimensional data, small sample sizes, and biopsy-inaccessible areas.
• ML-derived results represent associations, suitable for hypothesis generation, not causal confirmation.

Conclusions:

• Exploratory ML offers a complementary approach to traditional statistics in medical research.
• Evaluation of exploratory ML should prioritize stability, reproducibility, and interpretability over predictive performance.
• Adopting new evaluation standards can bridge exploratory analyses with confirmatory research and advance medical study designs.