Performance evaluation of machine learning and deep learning models for breast cancer detection

Abstract

Cancer of the breast remains a critical global health challenge, and as such, accurate and early detection is crucial for better patient management, and as such enhanced survival outcomes. Deep learning models have demonstrated exceptional performance in medical image analysis, traditional machine learning approaches using structured clinical data remain very effective and computationally efficient. In this study, a parallel evaluation of deep learning and machine learning models for breast cancer detection using clinical and mammographic datasets is presented. Clinical records from 600 patients were analyzed using Support Vector Machines, Decision Tree classifiers, and Naïve Bayes, while a total of 1,375 mammograms were analyzed using MobileNetV2, EfficientNet B0, XceptionNet, and VGG16 architectures. A five-fold cross-validation approach was employed, and performance was evaluated using F1-score, recall, precision, and accuracy.  Experimental results show that machine learning models achieved better performance, with Decision Tree and Support Vector Machine attaining accuracies of 97.67% and 97.30%, respectively, as opposed to a maximum of 74.18% achieved by XceptionNet among deep learning models. These results underscore the high predictive strength of structured clinical data and highlight the importance of data modality in model selection, while supporting the need for multi-modal and explainable AI frameworks for improved diagnostic systems in the future.