Prediction of Breast Cancer using Machine Learning Approaches

Area of Science:

• Oncology
• Medical Informatics
• Data Science

Background:

• Breast cancer is a prevalent cancer in women, influenced by diverse clinical, lifestyle, social, and economic factors.
• Machine learning (ML) offers a promising approach for breast cancer prediction by identifying complex patterns within data.

Purpose of the Study:

• To predict breast cancer using various machine learning techniques.
• To evaluate the impact of demographic, laboratory, and mammographic data on prediction accuracy.

Main Methods:

• An analytical study utilized a dataset of 5,178 records from Motamed Cancer Institute, with 25% of cases being breast cancer patients.
• Four machine learning algorithms were employed: Random Forest (RF), Multilayer Perceptron (MLP), Gradient Boosting Trees (GBT), and Genetic Algorithms (GA).
• Models were trained using demographic and laboratory features (20 features) and subsequently with all features including mammography (24 features) to assess mammography's contribution.

Main Results:

• Random Forest (RF) demonstrated superior performance with 80% accuracy, 95% sensitivity, 80% specificity, and an Area Under the Curve (AUC) of 0.56.
• Gradient Boosting (AUC=0.59) outperformed the neural network in predictive capability.
• The inclusion of mammographic features was evaluated for its effectiveness in enhancing breast cancer prediction.

Conclusions:

• Integrating multiple risk factors into predictive models can facilitate early breast cancer diagnosis and inform care planning.
• Intelligent systems leveraging diverse data types are crucial for effective breast cancer management and early detection.