Evaluating the predictive performance of subtyping: A criterion for cluster mean-based prediction

Area of Science:

• Data analysis across medicine, biology, and social sciences.
• Statistical modeling and machine learning applications.

Background:

• Heterogeneity in population data presents challenges for accurate analysis.
• Current subtyping methods lack robust validation metrics.
• Clustering algorithms are commonly used to address heterogeneity by creating homogeneous subgroups.

Purpose of the Study:

• To propose a novel metric for evaluating the predictive accuracy of subtyping methods.
• To introduce the cost of cluster mean-based prediction (CCMP) as a quantitative measure for subtype validation.
• To enable the selection of optimal subtype classifications based on prediction accuracy.

Main Methods:

• Development of the cost of cluster mean-based prediction (CCMP) metric.
• Application of CCMP to evaluate and compare different candidate clustering results.
• Validation of CCMP's computational implementation through numerical experiments.

Main Results:

• CCMP provides a reliable method for assessing prediction accuracy derived from subtyping.
• Selecting the minimum CCMP value identifies the optimal subtype classification.
• Numerical experiments confirmed the validity and utility of the CCMP metric.

Conclusions:

• The cost of cluster mean-based prediction (CCMP) offers a robust approach to validating subtype classifications.
• CCMP facilitates the selection of superior subgrouping strategies for enhanced predictive modeling.
• This metric addresses a critical need for evaluating subtyping validity in heterogeneous datasets.