Dynamic Coati Optimization Algorithm for Biomedical Classification Tasks.

Area of Science:

• Computer Science
• Machine Learning
• Bioinformatics

Background:

• Medical datasets often contain redundant and irrelevant features, increasing dimensionality.
• High dimensionality negatively impacts machine learning classifier performance, particularly for algorithms like k-Nearest Neighbors (kNN).
• Feature selection is a crucial technique to address these challenges in medical data analysis.

Purpose of the Study:

• To propose a novel dynamic feature selection method to improve the performance of medical diagnostic classifiers.
• To enhance the k-Nearest Neighbors (kNN) classifier by selecting essential attributes from medical datasets.
• To introduce the dynamic Coati Optimization Algorithm (DCOA) for efficient and adaptive feature selection.

Main Methods:

• Developed a dynamic version of the Coati Optimization Algorithm (DCOA) incorporating dynamic opposing candidate solutions.
• Implemented DCOA as a feature selection technique where features are introduced iteratively during optimization.
• Evaluated DCOA against the original COA and seven other metaheuristic algorithms using the CEC'22 test suite and nine diverse medical datasets.

Main Results:

• The proposed DCOA demonstrated superior performance compared to seven well-known metaheuristic algorithms.
• Achieved an overall accuracy of 89.7%, with feature selection reducing data by 24%.
• Reported high performance metrics: sensitivity of 93.35%, specificity of 96.81%, and precision of 93.90%.

Conclusions:

• The dynamic Coati Optimization Algorithm (DCOA) is an effective feature selection method for medical datasets.
• DCOA enhances classifier performance without requiring parameter tuning, offering a practical advantage.
• The method shows significant potential for improving diagnostic accuracy in machine learning-based medical applications.