Yaşar, Şeyma2026-04-042026-04-0420252147-7892https://doi.org/10.33715/inonusaglik.1571883https://search.trdizin.gov.tr/tr/yayin/detay/1301694https://hdl.handle.net/11616/108168Pancreatic cancer is a highly lethal malignancy with poor prognosis and limited early diagnosis methods. In this study, 60 serum samples (30 pancreatic cancer patients, 30 controls) were analyzed to identify potential biomarkers for early detection using machine learning. Proteomic data were obtained via glycoprotein enrichment and mass spectrometry, identifying 232 proteins. After preprocessing, 29 proteins were selected using the Elastic Net method. XGBoost, optimized with 10-fold cross-validation, classified pancreatic cancer with high performance (AUC=0.850, accuracy=0.833). The SHAP method identified P02750 (Leucine-rich alpha-2-glycoprotein), P02766 (Transthyretin), P01031 (Complement C5), and P02649 (Apolipoprotein E) as key proteins affecting cancer risk. These biomarkers may play a crucial role in early diagnosis and personalized treatment, but further validation in larger studies is required. © 2025, Inonu University. All rights reserved.eninfo:eu-repo/semantics/openAccessBiomarkerExplainable artificial intelligencePancreatic cancerXGBoostArticle131627310.33715/inonusaglik.15718832-s2.0-105000849470Q41301694