Breast Cancer Prediction Using Fine Needle Aspiration Features and Upsampling with Supervised Machine Learning.

Area of Science:

• Oncology
• Biomedical Engineering
• Data Science

Background:

• Breast cancer is a leading cause of death in women, with increasing incidence worldwide.
• Current diagnostic methods require expert interpretation and can be time-consuming.
• There is a need for automated, accurate, and timely breast cancer detection systems.

Purpose of the Study:

• To develop an automated system for accurate breast cancer prediction.
• To identify optimal features from fine needle aspiration (FNA) data for enhanced prediction.
• To evaluate the efficacy of various feature selection techniques and classifiers.

Main Methods:

• Feature selection techniques including Principal Component Analysis (PCA), Singular Vector Decomposition (SVD), and Chi-square (Chi2) were applied.
• Experiments were conducted with varying feature sets and sizes.
• The Synthetic Minority Over-sampling Technique (SMOTE) was used to address data imbalance.
• Six classifiers (Random Forest, SVM, Gradient Boosting, Logistic Regression, MLP, KNN) were tuned and evaluated.

Main Results:

• K-nearest neighbors (KNN) demonstrated superior performance.
• 100% accuracy was achieved using KNN with 20 features selected by SVD.
• High accuracy was also obtained with KNN using the 15 most important features identified by PCA.

Conclusions:

• Automated feature selection and classification significantly enhance breast cancer prediction accuracy.
• The proposed approach, particularly using KNN with PCA or SVD features, offers a promising solution for early and accurate breast cancer diagnosis.
• This research contributes to overcoming the accuracy gap in automated cancer detection systems.