Research on eight machine learning algorithms applicability on different characteristics data sets in medical...

Area of Science:

• Data Science
• Medical Informatics
• Machine Learning

Background:

• The rapid expansion of data mining necessitates efficient algorithm selection, particularly for specialized fields like medicine.
• Medical professionals face challenges in choosing appropriate data mining algorithms for diverse medical datasets.
• Existing methods lack tailored guidance for matching algorithms to specific medical data characteristics.

Purpose of the Study:

• To analyze and compare characteristics of medical datasets against general datasets from other fields.
• To establish rules for selecting data mining algorithms based on medical data attributes.
• To enhance the applicability of data mining techniques in medical research.

Main Methods:

• Quantified medical dataset characteristics using 26 indicators (simple, statistical, information theory).
• Selected eight representative machine learning algorithms based on maturity, usability, and family diversity.
• Evaluated algorithm performance via prediction accuracy, running speed, and memory consumption.

Main Results:

• Developed decision tree and stepwise regression models to learn algorithm applicability based on dataset metadata.
• Achieved over 75% accuracy in predicting algorithm applicability through cross-validation.
• Generated a knowledge base for selecting appropriate data mining algorithms for medical datasets.

Conclusions:

• The developed models demonstrate the validity and feasibility of algorithm applicability knowledge for medical data.
• This research provides a practical framework for optimizing data mining algorithm selection in the medical field.
• Improved algorithm selection can lead to more efficient and effective medical data analysis and insights.