Qluster: An easy-to-implement generic workflow for robust clustering of health data

Area of Science:

• Data Science
• Biostatistics
• Clinical Research

Background:

• Clustering algorithms enhance health data exploration by identifying sub-profiles within populations.
• Standardization of data science methods in clinical research lags behind traditional biostatistics, leading to variability in project execution.
• Lack of standardized approaches impacts algorithm choice, reliability, and credibility in data science projects.

Purpose of the Study:

• To propose Qluster, a practical and standardized workflow for performing clustering tasks in health data analysis.
• To offer a compromise between application genericity, ease of implementation, and robustness in data clustering.
• To provide a reliable solution for both novice and experienced data scientists for routine data mining.

Main Methods:

• The Qluster workflow emphasizes parsimonious and judicious selection of algorithms and implementations.
• It is designed for genericity, applicable to small/big data, mixed variable types, and high-dimensionality.
• Robustness is ensured through proven algorithms, stable cluster evaluation, and management of noise and multicollinearity.

Main Results:

• The proposed workflow offers a standardized, robust, and easy-to-implement solution for data clustering.
• Qluster can be automated and routinely applied to various clustering projects.
• A detailed application on a concrete use case demonstrates its practical utility.

Conclusions:

• Qluster provides a valuable tool for improving the standardization and reliability of data science in clinical research.
• It facilitates easier and more robust data clustering for less experienced data scientists.
• Experienced data scientists can leverage Qluster for straightforward and dependable preliminary data mining.