Machine learning models typically require labeled data for training, but the labeling process can be time-consuming. Unsupervised learning techniques, such as clustering, offer an attractive alternative for data labeling by extracting relevant information without relying on ground truth. In this study, we applied clustering methods to categorize breast cancer patients into benign and malignant groups based on their features, developing a novel pipeline for this task. The dataset, which includes 2,435 unclassified patient cases, was obtained from different medical laboratories from Sebha city, Libya. We evaluated six clustering methods: K-Means, K-Medoids, DBSCAN, Agglomerative Hierarchical, Gaussian-Mixture, and Mini-Batch K-Means. Gaussian-Mixture emerged as the top performer based on both internal cluster validity and external agreement metrics. It successfully partitioned the data into two categories, which were subsequently used to train an Artificial Neural Network (ANN). The ANN demonstrated remarkable accuracy, recall, precision, and F1-score of 99%, 99%, 97%, and 98%, respectively, when predicting malignant samples on a held-out test set of 609 samples. This high accuracy is of particular significance; as precise diagnosis is crucial for effective breast cancer treatment. The model was further deployed at the Sebha Oncology Center to provide a diagnostic service, where it demonstrated consistent performance, correctly classifying 99% of malignant samples. These results highlight the efficacy of the proposed novel techniques for breast cancer classification.