Machine learning models in breast cancer survival prediction

Area of Science:

• Oncology
• Biomedical Informatics
• Machine Learning in Healthcare

Background:

• Breast cancer poses a significant global health challenge with high mortality rates.
• Early diagnosis dramatically improves survival rates, increasing them from 56% to over 86%.
• Accurate predictive models are essential for timely and effective breast cancer management.

Purpose of the Study:

• To develop and evaluate a hybrid rule-based and machine learning model for predicting breast cancer survival.
• To compare the performance of various machine learning algorithms in breast cancer survival prediction.

Main Methods:

• Utilized a dataset comprising 900 patient records (97.3% female) with eight attributes.
• Applied and compared seven machine learning techniques: Naive Bayes (NB), Trees Random Forest (TRF), 1-Nearest Neighbor (1NN), AdaBoost (AD), Support Vector Machine (SVM), RBF Network (RBFN), and Multilayer Perceptron (MLP).
• Employed a 10-cross fold validation technique and evaluated performance using accuracy, precision, sensitivity, specificity, and area under the ROC curve.

Main Results:

• Trees Random Forest (TRF) outperformed other models, achieving 96% accuracy, 96% sensitivity, and 0.93 area under the ROC curve.
• Naive Bayes (NB), 1-Nearest Neighbor (1NN), AdaBoost (AD), Support Vector Machine (SVM), RBF Network (RBFN), and Multilayer Perceptron (MLP) showed varying performance levels.
• 1NN exhibited the poorest performance with 91% accuracy, 91% sensitivity, and 0.78 area under the ROC curve.

Conclusions:

• The rule-based Trees Random Forest (TRF) model is highly accurate for breast cancer survival prediction.
• TRF is recommended as a reliable tool to aid medical decision-making in oncology.
• Machine learning models can significantly assist clinicians in reducing diagnostic errors and improving patient care.