Machine learning (ML) techniques to predict breast cancer in imbalanced datasets: a systematic review

Area of Science:

• Medical Informatics
• Machine Learning
• Data Mining

Background:

• Knowledge Discovery in Databases (KDD) is vital for medical outcome prediction, especially for high-impact diseases like cancer.
• Machine learning (ML) offers potential for accurate cancer prediction but struggles with imbalanced datasets, leading to bias and overfitting.
• Breast cancer prediction is a critical area where imbalanced data presents significant challenges for developing reliable models.

Purpose of the Study:

• To systematically review machine learning methods used in breast cancer prediction from 2008-2023.
• To examine ML techniques addressing imbalanced data across KDD steps: preprocessing, data mining, and interpretation.
• To synthesize research on effective ML strategies for accurate breast cancer outcome prediction.

Main Methods:

• Systematic literature review of breast cancer prediction articles (2008-2023).
• Analysis of machine learning techniques applied in preprocessing (balancing, feature selection), data mining, and interpretation.
• Focus on methods designed to mitigate the impact of imbalanced datasets.

Main Results:

• Identification of effective preprocessing strategies, including data balancing and feature selection techniques.
• Synthesis of robust machine learning models and appropriate evaluation metrics for imbalanced breast cancer data.
• Overview of ML approaches that enhance the accuracy and reliability of cancer prediction models.

Conclusions:

• Effective ML preprocessing and modeling techniques are crucial for overcoming imbalanced data challenges in breast cancer prediction.
• This review provides insights into robust methods for healthcare providers and researchers to improve diagnostic accuracy.
• Informed application of ML can lead to better healthcare decisions and patient outcomes in oncology.