An automated process for supporting decisions in clustering-based data analysis.

Area of Science:

• Biomedical Informatics
• Data Science
• Quantitative Research Methods

Background:

• Biomedical researchers frequently use quantitative metrics to evaluate diverse elements, including datasets and models.
• A lack of standardized validation procedures for metrics can lead to assumptions about their generalizability across datasets.
• The behavior of metrics can vary significantly across different scenarios, challenging their universal applicability.

Purpose of the Study:

• To investigate and assess the behavior of quantitative metrics within biomedical research contexts.
• To develop a framework for evaluating metric reliability before applying them to biomedical datasets.
• To enhance decision-making processes for selecting appropriate metrics for specific analytical tasks.

Main Methods:

• A novel method employing clustering-based data analysis to evaluate quantitative metric behavior.
• Assessment of metrics using unsupervised classification validation criteria: cluster stability and goodness.
• Development of the evaluomeR tool to facilitate the application of this method for biomedical researchers.

Main Results:

• Demonstrated the analytical power of the proposed method through diverse applications.
• Analyzed the behavior of the impact factor metric across various journal categories.
• Identified structural metrics for optimal partitioning of biomedical ontology repositories.
• Investigated sources of heterogeneity in effect size metrics from biomedical studies.

Conclusions:

• Statistical properties like stability and goodness of classification offer valuable insights into metric behavior.
• The proposed method supports informed decisions regarding the selection of appropriate metrics for specific biomedical datasets.
• Reliable metric evaluation is essential for accurate conclusions drawn from biomedical data analysis.