Comparing Machine Learning Techniques for Breast Cancer Prediction: A Comprehensive Study

This study conducted a comprehensive comparative analysis of machine learning (ML) techniques for predicting breast cancer risk, using the Breast Cancer Wisconsin (Di-agnostic) dataset. We explored the efficacy of 10 ML models and 3 feature selection methods along with hyperparameter optimization. Our primary focus of performance is recall, accuracy, and F1 measures. We obtained our metrics from cross validation for realistic performance predictions. ML models included support vector, decision tree, logistic regression, k-nearest neighbors, random forest, Gaussian Naïve Bayes, and 4 flavors of boosting ensemble algorithms. Feature selection techniques included univariate, recursive selections and principal component analysis. Our analysis provides valuable and quickly attainable insights into the practical application of ML techniques in clinical settings for breast cancer screening.