Leveraging Synthetic Minority Over-Sampling Technique for Class Imbalance in Machine Learning-based Breast Cancer Diagnosis

In this research, we conducted a comprehensive analysis of the diagnosis of breast cancer tumors as benign or malignant employing machine learning algorithms. The publicly available breast cancer diagnosis dataset was taken from the UCI repository of 569 breast cancer patients. The class imbalance in the dataset of 63% benign and 37% malignant cases was first handled by employing the synthetic minority oversampling technique. Then, a comparative analysis of six different machine-learning algorithms was done. The performance evaluation of all six models showed that the logistic regression model outperforms the other models. We evaluated it with recall, precision, and f1-score which was found to be 98.68%, 97.40%, and 97.90%, respectively. To ensure the scalability and generalization of the logistic regression model for the breast cancer dataset, a learning curve was obtained, which offered a reliable and interpretable solution for the classification of breast cancer tumors.