Clinical data classification using an enhanced SMOTE and chaotic evolutionary feature selection.

Area of Science:

• Machine Learning
• Medical Informatics
• Data Science

Background:

• Class imbalance and irrelevant features hinder classification model development.
• Clinical Decision Support Systems (CDSS) require robust methods to handle complex datasets.
• Existing methods may not adequately address both data imbalance and feature selection simultaneously.

Purpose of the Study:

• To propose and evaluate a novel framework for developing CDSS that tackles class imbalance and feature selection.
• To enhance the performance of classification models in clinical settings.
• To improve the reliability and accuracy of medical diagnostic tools.

Main Methods:

• Dataset balancing using enhanced Synthetic Minority Over-sampling Technique (SMOTE) with Orchard's algorithm.
• Feature subset selection employing a wrapper approach with Chaotic Multi-Verse Optimisation (CMVO).
• Classification using a Random Forest (RF) classifier with Information Gain Ratio as the split criteria.

Main Results:

• The proposed framework achieved competitive performance across three clinical datasets (ILPD, TSD, PID).
• Performance metrics included Matthews correlation coefficient (MCC), F-score (F1), and accuracy, showing significant improvements.
• Statistical analysis (Wilcoxon test) confirmed the superiority of the proposed method over existing approaches.

Conclusions:

• The developed framework effectively addresses class imbalance and feature selection challenges in medical datasets.
• The integration of SMOTE, CMVO, and RF offers a powerful approach for building accurate CDSS.
• The proposed method demonstrates a statistically significant improvement in classification performance for clinical applications.