Enhancing generalization of machine learning models for oncology informatics

Generalization, the ability of machine learning models to perform well on new data, remains a major challenge in sensitive fields such as medical settings. This work addresses generalization issues in three preclinical oncology applications by leveraging standardized radiotherapy coding systems for learning from administrative records, introducing an ensemble uncertainty estimation method to enhance dataset comparison, and applying this method to improve automated skin cancer detection from images. These methods collectively aim to improve the reliability of AI in medical contexts, with future efforts focused on integrating generalization metrics into real-time clinical pipelines.