Evaluation of Feature Selection Techniques for Breast Cancer Risk Prediction.

Area of Science:

• Computational Biology and Bioinformatics
• Machine Learning in Healthcare
• Cancer Epidemiology

Background:

• Breast cancer is a significant public health concern, necessitating improved risk prediction models.
• Identifying crucial environmental and genetic factors is vital for accurate breast cancer risk assessment.
• Evaluating the stability and performance of feature selection methods is essential for robust predictive modeling.

Purpose of the Study:

• To identify the most important factors for breast cancer risk prediction using machine learning.
• To evaluate and compare the performance and stability of various feature selection techniques.
• To enhance the accuracy of breast cancer risk prediction models in a healthy population.

Main Methods:

• Utilized a dataset from the MCC-Spain study with 919 cases and 946 controls.
• Applied and assessed multiple feature ranking and selection algorithms (e.g., SVM-RFE, Random Forest, Relief, wrapper methods).
• Employed classifiers like Logistic Regression and evaluated model performance using the Area Under the ROC Curve (AUC) metric.

Main Results:

• The Support Vector Machine-Recursive Feature Elimination (SVM-RFE) ranking method demonstrated superior performance.
• Using the top 47 features selected by SVM-RFE with Logistic Regression yielded an AUC of 0.616, a 5.8% improvement over the full feature set.
• SVM-RFE and Random Forest exhibited high stability in feature selection, while Relief and wrapper methods were unstable.

Conclusions:

• SVM-RFE is a highly effective and stable feature selection technique for breast cancer risk prediction.
• Combining stable feature selection methods with robust classifiers significantly improves predictive model performance.
• The study highlights the importance of evaluating both stability and performance for reliable cancer risk assessment tools.