Feature selection using binary particle swarm optimization and support vector machines for medical diagnosis.

Area of Science:

• Medical Informatics
• Computational Biology
• Machine Learning

Background:

• Accurate medical disease diagnosis relies on effective feature selection.
• Traditional methods like F-score and information gain have limitations.
• Optimizing feature subsets is crucial for enhancing diagnostic model performance.

Purpose of the Study:

• To propose and evaluate a feature selection strategy using binary particle swarm optimization (BPSO) for medical disease diagnosis.
• To assess the efficacy of BPSO in improving diagnostic accuracy while reducing feature dimensionality.
• To compare the performance of the proposed BPSO method against traditional and genetic algorithm-based feature selection techniques.

Main Methods:

• A feature selection strategy employing binary particle swarm optimization (BPSO).
• Support vector machines (SVM) utilized as the fitness function within the BPSO algorithm.
• Evaluation conducted on four diverse medical datasets: SPECT heart, Wisconsin breast cancer, Pima Indians diabetes, and Dermatology.

Main Results:

• The BPSO method achieved higher diagnostic accuracy using a reduced number of features across all tested medical conditions (heart, cancer, diabetes, erythematosquamous diseases).
• Superior accuracy was observed when compared to F-score and information gain feature selection methods.
• The proposed BPSO approach demonstrated higher performance than a genetic algorithm in most cases, utilizing fewer features.

Conclusions:

• Binary particle swarm optimization offers a powerful and efficient feature selection strategy for medical disease diagnosis.
• The proposed method enhances diagnostic accuracy and reduces computational complexity by selecting optimal feature subsets.
• This approach shows significant potential for improving machine learning-based diagnostic systems in healthcare.