Patient-Centric Multilabel Classification with Svm on Breakhis Dataset

Histopathological image analysis is critical for early breast cancer detection. Traditional single-label classification models often overlook the complexity of clinical data, where images may exhibit multiple co-occurring features. This study presents a multilabel classification framework using the BreakHis dataset to address this challenge. DenseNet201 is used for deep feature extraction, and a One-vs-Rest Support Vector Machine (SVM) performs classification. Two strategies are proposed: Top-K prediction (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$k=2$</tex>) to manage diagnostic uncertainty, and hierarchical multilabel classification to capture structured label relationships. A patient-wise train-test split is adopted to avoid data leakage and ensure robust generalization. Experimental results show that the hierarchical strategy achieves 81.36 % accuracy, 67.72 % F1-score, and 85.48 % G-Mean, demonstrating its effectiveness in clinical scenarios. By aligning predictions with worst-case diagnostic outcomes, this approach enhances reliability in malignancy detection and supports more interpretable, patient-centric automated histopathological analysis.