The prevalence of pancreatic cancer has prompted the development of AI-driven diagnostic systems that can improve early detection and classification. This study presents a novel approach for pancreatic cancer classification using a decision tree model trained on genomic biomarkers derived from patient blood and urine samples. Contrary to traditional image-based methods, which typically result in limited sensitivity and specificity, this research leverages structured biological data for improved accuracy. The dataset, Pancrease_clean_data.csv, was sourced from Kaggle and contains clinically relevant features such as age, sex, CA 19-9, CEA, creatinine, LYVE1, REG1B, and TFF1. Data preprocessing included missing value imputation, normalization, and label encoding. The dataset was balanced using SMOTE (Synthetic Minority Over-sampling Technique) to address class imbalance. The class distribution post-balancing was 50:50 between cancer-positive and negative cases. Stratified sampling ensured consistent class proportions across training (80%) and test (20%) sets. The proposed decision tree model achieved an accuracy of 95.3%, outperforming traditional models in both recall and F1-score. This model demonstrates potential for clinical application, particularly in resource-constrained environments where imaging-based diagnostics may be impractical.