A novel multi-objective medical feature selection compass method for binary classification.

Area of Science:

• Artificial Intelligence in Medicine
• Computational Biology
• Medical Informatics

Background:

• Medical decision-making often involves complex, multi-objective optimization.
• Multi-Objective Feature Selection (MOFS) is a key challenge in combining AI and medicine.
• Existing methods may not adequately balance classification performance with feature set efficiency.

Purpose of the Study:

• To propose a novel MOFS approach for medical binary classification.
• To enhance the trade-off between the number of features, accuracy, and Area Under the ROC Curve (AUC).
• To ensure high classification quality for both healthy and ill patients.

Main Methods:

• Development of the Genetic Algorithm with multi-objective Compass (GAwC).
• Utilizing a Genetic Algorithm guided by a 3-Dimensional Compass.
• Evaluating the method on real-world medical datasets for binary classification tasks.

Main Results:

• GAwC demonstrated superior performance compared to other genetic algorithm-based MOFS approaches.
• The method effectively balances the number of features, accuracy, and AUC.
• GAwC ensures classification quality, crucial for accurate patient diagnosis.

Conclusions:

• GAwC offers a promising solution for MOFS in medical binary classification.
• The approach provides reliable classification quality by incorporating AUC as an objective.
• The method's effectiveness is validated on real-world medical data, with results discussed from medical and classification perspectives.