Breast Tumor Classification Using an Ensemble Machine Learning Method

Area of Science:

• Medical Informatics
• Machine Learning
• Artificial Intelligence

Background:

• Breast cancer remains a leading cause of mortality in women globally.
• Accurate and early detection of breast cancer is critical for improving patient outcomes.
• Artificial intelligence (AI) offers promising avenues for enhancing breast cancer diagnostic capabilities.

Purpose of the Study:

• To propose and evaluate an ensemble classification mechanism for breast cancer detection using AI.
• To compare the performance of various machine learning algorithms on the Wisconsin Breast Cancer Dataset (WBCD).
• To determine the optimal voting strategy for an ensemble model focused on minimizing false negatives.

Main Methods:

• Evaluated multiple state-of-the-art machine learning classifiers on the WBCD.
• Selected the top three classifiers based on the F3 score, prioritizing recall (minimizing false negatives).
• Implemented an ensemble classification using a majority voting mechanism (hard voting) with logistic regression, support vector machines, and multilayer perceptron networks.

Main Results:

• The ensemble classification model using hard (majority-based) voting achieved a high accuracy of 99.42%.
• This performance surpassed existing state-of-the-art algorithms on the WBCD.
• Different soft voting methods (average, product, max, min probabilities) were evaluated but showed lower performance than hard voting.

Conclusions:

• An ensemble classification approach, particularly with majority-based voting, demonstrates superior performance in breast cancer detection.
• The developed AI system offers a highly accurate method for breast cancer classification.
• This research highlights the potential of ensemble AI models in improving diagnostic accuracy for critical diseases like breast cancer.