[Feature extraction for breast cancer data based on geometric algebra theory and feature selection using differential...

Area of Science:

• Pattern Recognition
• Biomedical Data Analysis
• Machine Learning

Background:

• Feature extraction and selection are critical challenges in pattern recognition.
• High-dimensional data presents difficulties for traditional methods.
• Geometric algebra offers a novel representation for feature engineering.

Purpose of the Study:

• To propose a new feature extraction method using geometric algebra.
• To develop an improved differential evolution (DE) method for feature selection in high-dimensional datasets.
• To evaluate the efficacy of the proposed methods in biomedical data classification.

Main Methods:

• Geometric algebra-based feature extraction using blade coefficients.
• An enhanced differential evolution algorithm for feature selection.
• Simple linear discriminant analysis (LDA) as the classifier.
• 10-fold cross-validation (10 CV) for performance evaluation.

Main Results:

• The proposed geometric algebra feature extraction achieved high performance.
• The improved DE feature selection effectively handled high-dimensional data.
• Classification accuracy on a public breast cancer dataset exceeded 96% using 10 CV.
• The combined approach demonstrated superiority over original features and traditional methods.

Conclusions:

• The proposed geometric algebra and DE-based feature engineering methods are effective for pattern recognition.
• These novel techniques offer significant improvements in biomedical data classification tasks.
• The methods show promise for addressing high-dimensional challenges in machine learning.