Uncontrolled cell growth in the pancreatic gland, is one of the most aggressive types of cancer with a high mortality rate, called pancreatic cancer. This research focuses on improving early diagnosis methods for pancreatic cancer by using CatBoost. Urine biomarker datasets were collected and subjected to pre-processing, including label coding, standardized scaling, and balancing via the Synthetic Minority Oversampling Technique (SMOTE). The CatBoost model achieved an accuracy of 98.89%, specificity of 99.35%, sensitivity of 98.71%, and Area Under the Curve (AUC) of 0.9951. These results show that the CatBoost model significantly outperforms the diagnosis models in previous studies, overcoming the challenges of early detection and classification of pancreatic cancer. This study shows that CatBoost is effective for diagnosing pancreatic cancer and suggests that future research explore other models on larger and more diverse datasets.