Enhancing Early Detection of Pancreatic Cancer: A Machine Learning Approach with Explainable AI Insights

Among the most deadly and complex illnesses in the world, pancreatic cancer is typified by its early asymptomatic stage and subsequent discovery of the disease when it has progressed to an advanced stage. Getting the patients in a correct treatment path at the earliest point is important for suitable intervention and patient outcomes in the long run. We utilized advanced algorithms and model prediction techniques includes, feature extraction, hyperparameter tuning and explainability methods which aims to enhance both the comprehension and predictive accuracy of models employed. The goal is to assess patient health status and diagnose pancreatic diseases (cancer vs. non-cancer) by developing a learning algorithm. We obtained a dataset containing demographic information and urine biomarkers to start the investigation. First stage of our analysis employs a single classifier and later we apply bagging and boosting algorithms to achieve ensemble models. Hyperparameter fine-tuning has got a positive outcome, which has Gradient Boosting and Cat Boosting as its top performing models. The ensemble model produced excellent results $(92.02 \%)$, demonstrating the usefulness of the suggested approach. LIME and SHAP techniques play the key role in model outcome interpretability providing an insight of algorithmic decision-making process. Such a wide range of combinations of biomarkers have a potential of pancreatic cancer detection in early stages and thus will help shape diagnostics strategies in near future.