Breast cancer is among the most prevalent cancers that affect women worldwide, and successful treatment and improved patient outcomes depend on early and accurate detection. Medical imaging has greatly improved in recent years thanks to deep learning-based techniques, which provide high-accuracy solutions thanks to their layered structures. However, existing deep learning-based breast cancer classification models still face challenges such as inaccurate disease classification and the unavailability of large, diverse datasets required to enhance diagnostic efficacy. This research introduces a deep learning-based classification model designed to improve the accuracy of breast cancer detection. Initially, mammogram images are gathered from publicly available. A new deep learning architecture called Attention-based Multi- View Swin Patch Transformer (Att-MVSPtrans) was created for the early classification of breast cancer. This model incorporates attention mechanisms to focus on important image regions, a patch embedding technique for efficient processing, and residual layers for improved feature learning and propagation. Accuracy, Precision, Recall, F1-Score, and False Positive Rate (FPR) values of 0.9938, 0.9915, 0.9914, 0.9912, and 0.0059, respectively, show that the model is effective in accurately and early breast cancer classification. © 2025 Elsevier B.V., All rights reserved.