A modified and weighted Gower distance-based clustering analysis for mixed type data: a simulation and empirical...

Area of Science:

• Data Science
• Biostatistics
• Computational Biology

Background:

• Traditional clustering methods struggle with mixed-type data (continuous and categorical variables).
• Real-world clinical datasets frequently contain a mix of data types, necessitating advanced analytical approaches.
• Existing techniques lack adaptability and interpretability for complex, heterogeneous datasets.

Purpose of the Study:

• To introduce a novel clustering technique, DAFI (Distance-based Algorithm with Feature Importance), designed for mixed-type data.
• To enhance clustering compatibility, adaptability, and interpretability in clinical research.
• To identify distinct patient clusters and their associated health profiles using real-world data.

Main Methods:

• Developed a modified Gower distance metric incorporating feature importance weights.
• Evaluated the DAFI algorithm on simulated datasets and the National Health and Nutrition Examination Survey (NHANES) data.
• Compared DAFI's performance against 13 existing clustering techniques using Adjusted Rand Index (ARI) and silhouette scores.

Main Results:

• DAFI consistently outperformed baseline methods in simulation studies, particularly with redundant features (ARI).
• DAFI achieved the highest silhouette score (0.79) on NHANES data, identifying four distinct health clusters.
• Feature importance highlighted cardiovascular disease (CVD) risk factors in cluster formation.
• Analysis revealed a significant association between periodontitis (PD) and CVDs (adjusted OR 1.95).

Conclusions:

• DAFI demonstrates superior performance over traditional clustering baselines for both simulated and real-world mixed-type data.
• The algorithm effectively captures cluster characteristics by weighting feature importance, crucial for clinical relevance.
• DAFI provides a robust and interpretable solution for mixed-type data clustering in healthcare settings.