A Comparative Study of Oversampling Techniques for Predicting Breast Cancer Recurrence

Breast cancer recurrence prediction is crucial for patient care, necessitating advanced methodologies to enhance precision and facilitate therapeutic decision-making. This paper evaluates how oversampling techniques affect breast cancer recurrence prediction on the base classifiers. Our study is based on the Jordan Breast Cancer Dataset (JBRCA), which contains 21 features and 9723 instances. The JBRCA is a breast cancer dataset extracted from the King Hussein Cancer Center's (KHCC) registry database. By illustration, the study describes the imbalance problem facing most real datasets. This imbalance can lead to low model performance. Over-sampling techniques can help balance the data when there are fewer recurrence cases than non-recurrence. The proposed methodology includes data preparation and involves oversampling techniques for model development. We used base classification models such as logistic regression, decision tree, k-nearest neighbors, Gaussian Naive Bayes, and multilayer perceptron, combining them with oversampling techniques SMOTE, Random, ADASYN, SVMSMOTE, and Borderline-SMOTE to evaluate the compiled dataset. We examine the impact of the base classifiers along with oversampling techniques using accuracy, sensitivity, specificity, G-mean, and ROC-AUC to assess the model's efficacy.