Recursive Statistical Invariants Elimination for High-Risk Disease Diagnosis With Feature Selection

Area of Science:

• Biomedical Informatics
• Machine Learning in Healthcare
• Computational Biology

Background:

• Accurate feature selection is crucial for high-risk disease diagnosis and risk assessment.
• Existing methods often neglect medical domain knowledge, potentially compromising diagnostic accuracy.

Purpose of the Study:

• To introduce Recursive Statistical Invariants Elimination (RSIE), a novel framework for feature selection in high-risk disease diagnosis.
• To systematically integrate medical domain knowledge into the feature selection process.

Main Methods:

• RSIE combines Learning Using Statistical Invariants (LUSI) for pattern extraction, kernel alignment for knowledge-dataset correlation, and Recursive Feature Elimination (RFE) for selection.
• The framework was evaluated on 15 high-risk disease datasets (cardiovascular, neurological, oncological).

Main Results:

• RSIE achieved an average accuracy of 88.74%, a 5.46% improvement over traditional SVM-RFE (83.28%).
• The LUSI-RFE* variant reduced feature dimensions by 45%-74% while preserving high classification accuracy.
• RSIE demonstrated stability under high noise, computational efficiency, and improved model interpretability.

Conclusions:

• RSIE effectively integrates medical domain knowledge for superior feature selection in high-risk disease diagnosis.
• The proposed method offers a robust, efficient, and interpretable solution for improving diagnostic accuracy.