A genetic algorithm based nearest neighbor classification to breast cancer diagnosis

Area of Science:

• * Computational intelligence
• * Medical informatics
• * Biomedical engineering

Background:

• * Accurate breast cancer diagnosis is crucial for effective treatment, yet challenging even for experts.
• * Automatic classification systems are highly desirable to support oncologists in differentiating benign and malignant tumors.
• * Existing methods may require significant computational resources and feature optimization.

Purpose of the Study:

• * To present a hybrid genetic algorithm (GA) and k-nearest neighbor (KNN) approach for breast tumor classification.
• * To develop an automated system for classifying benign and malignant breast tumors using Wisconsin breast cancer data.
• * To optimize the classification process by minimizing reference patterns and features while maximizing performance.

Main Methods:

• * Application of a hybrid GA-KNN algorithm for breast cancer data classification.
• * Utilizing GA for selecting a compact reference set from training patterns in KNN.
• * Simultaneous feature selection and pattern selection by GA to prune unnecessary data.

Main Results:

• * The GA-KNN method demonstrated effective classification of benign and malignant breast tumors.
• * The approach successfully reduced the number of reference patterns and features.
• * Performance was compared against a fuzzy-genetic approach, indicating the hybrid method's viability.

Conclusions:

• * The hybrid GA-KNN approach offers a promising tool for automated breast cancer diagnosis.
• * This method enhances classification efficiency by optimizing pattern and feature selection.
• * The study contributes to developing advanced decision support systems for oncologists.