Answers are grounded in this paper and cite their sources.
Class imbalance remains a persistent challenge in medical diagnosis, particularly in breast cancer datasets, which often leads to models that overlook malignant cases. While the Synthetic Minority Oversampling Technique (SMOTE) and its variants address this issue, they frequently introduce noise and fail to preserve the intrinsic data structure. This study proposes a sequentially integrated resampling framework combining Adaptive Synthetic Sampling (ADASYN), Cluster-based oversampling and density-based filtering to enhance data quality and classification performance. The framework was evaluated using extreme gradient boosting (XGBoost) on the Wisconsin Diagnostic Breast Cancer (WDBC) dataset, which contains 569 observations and 30 real-valued diagnostic features derived from digitized images of fine-needle aspirates (FNA) of breast masses. The original data also contains an identification variable and a diagnosis variable. The identification variable was excluded from model development because it has no diagnostic meaning, while the diagnosis variable was retained as the target variable and encoded as malignant = 1 and benign = 0. Thus, only the 30 continuous diagnostic features were used as predictor variables. The dataset was split into training and testing sets. The proposed improved SMOTE achieved an accuracy of 96.49%, recall of 97.22%, precision of 97.22%, F1 score of 97.22% and area under the receiver operating characteristic curve (ROC-AUC) of 96.23%, achieving higher threshold-dependent classification performance while yielding a lower ROC-AUC than the evaluated baseline resampling approaches using the same XGBoost classifier.
No takes yet. Share an insight, caveat, or question.
Kutigi et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: