Evaluating Auto Machine Learning Strategies for Breast Cancer Classification Using the Wisconsin Breast Cancer Diagnostic Dataset: Comparative Study
Year: 2026
Type: article
Abstract: Objective: To compare three automated machine learning (AutoML) frameworks for benign--malignant breast cancer classification using the Wisconsin Breast Cancer Diagnostic (WBCD) dataset, focusing on performance, stability, algorithm selection, and explainability. Material and Methods: Tree-based Pipeline Optimization Tool (TPOT), Fast and Lightweight AutoML (FLAML), and Machine Learning Jar (MLJAR) were evaluated with 50 repeated stratified hold-out splits. In each repetition, 80% of the data were used for training and 20% for testing while preserving class distribution. Performance was assessed by accuracy, sensitivity, specificity, F1- score, area under the receiver operating characteristic curve (AUCROC), and area under the precision-recall curve (PR-AUC). Algorithm-selection frequencies were recorded, and post-hoc explainability was examined using SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME). Results: All frameworks achieved strong discrimination. FLAML showed the highest accuracy (0.976±0.016), sensitivity (0.986±0.018), F1-score (0.981±0.013), and AUC-ROC (0.994±0.006), while MLJAR achieved the highest PR-AUC (0.995±0.008). TPOT retained high sensitivity but showed lower specificity and greater variability. FLAML and MLJAR produced more stable performance across repeated partitions. In the representative explainability analysis, SHAP and LIME highlighted worst concave points, worst perimeter, and worst radius as influential predictors. Conclusion: AutoML frameworks can achieve strong internal benchmark performance on structured breast cancer classification data, but their practical value should be judged together with stability, search behaviour, explainability, and resource requirements.
Institution: Sakarya University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7202287665