Feature selection using genetic algorithm for breast cancer diagnosis: experiment on three different datasets.

Area of Science:

• Biomedical Informatics
• Machine Learning in Healthcare
• Computational Biology

Background:

• Accurate breast cancer diagnosis is crucial for effective treatment.
• Feature selection is vital for optimizing diagnostic model performance.
• Wisconsin breast cancer datasets are widely used benchmarks.

Purpose of the Study:

• To evaluate a wrapper-based feature selection method for breast cancer diagnosis.
• To compare the effectiveness of different classifiers with and without feature selection.
• To assess the impact of feature selection on accuracy, sensitivity, and specificity.

Main Methods:

• Employed a Genetic Algorithm (GA)-based wrapper approach for feature selection.
• Utilized three classifiers: Artificial Neural Network (ANN), PS-classifier, and GA-classifier.
• Tested the methods on Wisconsin breast cancer (WBC), Wisconsin diagnosis breast cancer (WDBC), and Wisconsin prognosis breast cancer (WPBC) datasets.

Main Results:

• Feature selection improved accuracy for most classifiers across datasets.
• The PS-classifier achieved the best accuracy on the WBC dataset with feature selection.
• The ANN achieved the best accuracy on the WDBC and WPBC datasets with feature selection, alongside improved specificity and sensitivity.

Conclusions:

• Feature selection demonstrably enhances classifier accuracy, specificity, and sensitivity in breast cancer diagnosis.
• The proposed GA-based feature selection method shows competitive performance compared to existing studies.
• The optimal classifier performance post-feature selection varies depending on the specific Wisconsin breast cancer dataset used.