A novel double machine learning approach for detecting early breast cancer using advanced feature selection and...

Area of Science:

• Medical Informatics
• Machine Learning
• Computational Biology

Background:

• Accurate breast cancer detection is crucial for effective treatment and patient outcomes.
• Traditional machine learning methods may struggle with complex, high-dimensional datasets.
• Integrating diverse data types and learning approaches can improve diagnostic accuracy.

Purpose of the Study:

• To propose and evaluate three novel Double Machine Learning (DML) models for enhanced breast cancer detection.
• To leverage the strengths of different machine learning and deep learning algorithms through feature fusion.
• To achieve high classification performance in breast cancer detection using a comprehensive dataset.

Main Methods:

• Developed three DML models combining established ML algorithms (Random Forest, XGBoost, LightGBM) with deep learning (Feedforward Neural Network, Artificial Neural Network).
• Employed feature fusion techniques where outputs from base models are integrated by a meta-classifier for final predictions.
• Incorporated dimensionality reduction (Principal Component Analysis) and feature selection to optimize model performance.

Main Results:

• All three proposed DML models demonstrated significant improvements in breast cancer detection accuracy.
• The models effectively integrated structured, non-linear, and sequential data features.
• An impressive accuracy of 0.99 was achieved, highlighting the efficacy of the proposed DML approaches.

Conclusions:

• The developed DML models offer a powerful framework for accurate breast cancer detection.
• Feature fusion and the combination of diverse ML/DL techniques are key to enhancing diagnostic performance.
• These models hold promise for advancing automated breast cancer screening and diagnosis systems.