Breast Cancer Prediction Using Rotation Forest Algorithm Along with Finding the Influential Causes.

Area of Science:

• Biomedical Engineering
• Machine Learning in Healthcare
• Oncology

Background:

• Breast cancer (BC) is a significant global health concern, primarily affecting women.
• Early detection and accurate classification are crucial for improving patient outcomes.
• Machine learning offers promising avenues for enhancing BC diagnosis and risk assessment.

Purpose of the Study:

• To classify breast cancer (BC) versus non-BC cases using machine learning.
• To optimize classifier performance and feature selection for improved diagnostic accuracy.
• To identify key influential features for BC classification using counterfactual explanations.

Main Methods:

• Utilized the Rotation Forest classifier with hyperparameters optimized by the Optuna optimizer.
• Employed Sequential Forward Selection, Sequential Backward Selection, and Exhaustive Feature Selection for feature selection.
• Developed an ensemble model with soft and hard voting strategies for classification.

Main Results:

• The hard voting strategy achieved superior performance with 85.71% accuracy, 83.87% F1-score, 92.85% precision, and 76.47% recall.
• The soft voting strategy yielded 80.00% accuracy, 77.42% F1-score, 85.71% precision, and 70.59% recall.
• Diverse Counterfactual Explanations identified BMI and Glucose as highly influential features, while HOMA, Adiponectin, and Resistin showed minimal impact.

Conclusions:

• The optimized Rotation Forest classifier with a hard voting ensemble significantly improves breast cancer classification accuracy.
• BMI and Glucose levels are critical indicators for predicting breast cancer, highlighting their importance in clinical assessment.
• Feature selection and hyperparameter optimization are essential for maximizing the effectiveness of machine learning models in oncology.