Pancreatic Cancer Prediction with LightGBM Feature‑Selected classifier, Optimized by Sea Lion Metaheuristic Algorithm

Pancreatic cancer is one of the deadliest cancers due to late diagnosis. Typically, discovered after the advanced stages when there is no longer any hope of effective medical treatment. Thus, timely detection via automated approaches is critical for improved diagnostic and treatment outcomes. As the primary study goal, we create a robust machine learning method, specifically Light Gradient Boosting Machine (LightGBM), that incorporates a metaheuristic approach of feature selection based on CA۱۹-۹ and urinary biomarkers to accomplish early pancreatic cancer detection with optimum predictive performance. To find the most significant biomarkers, we emphasize the need to use the Sea Lion Optimization (SLO) method as a powerful feature selection strategy. Additionally, we employ Grid Search to optimize hyperparameters, further enhancing model performance. This model constantly outperforms individual classifiers in crucial metrics such as accuracy, precision, recall, and F۱ score. On this dataset, the model achieves ۰.۹۷ accuracy, ۰.۹۳ recall, ۰.۹۵ precision, and ۰.۹۶ F۱ score. This demonstrates its ability to detect positive cases and maintain balanced performance. The results demonstrate that optimized LightGBM models, combined with effective feature selection. This technique improves model performance and also provides insights into the main aspects of pancreatic cancer diagnosis. It has the potential to help clinicians with early detection and tailored treatment planning.

Keywords:

Pancreatic cancer Diagnosis , Metaheuristic Optimization , LightGBM , Sea Lion Optimization