International Journal of Scientific Research and Engineering Development

Spotting breast cancer early saves lives. That is the bottom line. It also takes a massive weight off the shoulders of overworked hospital staff. We wanted to build something that actually helps with this, so we put together a new model called XG-GATNet. It is a Graph Attention Network tailored specifically for classifying breast tumors. We tested it out using the well-known Wisconsin Diagnostic Breast Cancer (WDBC) dataset. Instead of just treating clinical records as a flat spreadsheet, our model maps out the features as a network of connected nodes, paying close attention to the data points that actually matter. We did not want to just test it in a vacuum, though. We stacked XG-GATNet up against some heavy hitters: Logistic Regression, XGBoost, an MLP, and GraphSAGE. We made sure the playing field was level by applying Z-score normalization and stratified 5-fold cross-validation across the board. The results actually exceeded our expectations. XG-GATNet beat all the baselines, landing the highest overall accuracy at 98.24%. It also held steady with a 97.61% precision, recall, and F1-score. Sure, GraphSAGE gave us a 100% precision rate, and the MLP was right behind with a 97.72% accuracy. But XG-GATNet clearly won out by figuring out the deep structural relationships hiding in the data. Basically, using graph connectivity on structured tabular records is a massive leap forward. It could seriously improve how computer-aided diagnostic systems run in the real world.

Harinarayanan R, Praveen Kumar G, Senthil Dheeraj V, Baskaran G, Abishek G, Balamurali M,"XG-GATNet: Modeling Structural Feature Relationships for Highly Accurate Breast Cancer Diagnostics" International Journal of Scientific Research and Engineering Development, V9(1): Page(2180-2187) Jan-Feb 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.