Cancer is characterized by a set of well-studied and recognized hallmarks. At the very early stages of cancer, subtle changes occur at the cellular and molecular levels long before tumors become clinically evident. These early indicators, such as epigenetic alterations, biomolecular patterns, and imaging markers, precede the development of hallmark features associated with cancer.

Machine learning (ML) is transforming early cancer diagnosis by leveraging mathematical models to uncover patterns beyond human detection. By analyzing large and complex datasets, these models can predict cancer risk well before traditional methods can detect any signs. This early prediction capability allows ML models to identify specific signatures associated with cancer development. However, translating these predictions into clinical practice requires explainability to ensure that medical professionals can interpret and trust the findings.

This presentation will explore some of the mathematical methods used in explainable ML, such as feature selection and gradient boosting algorithms. Gradient boosting is an ensemble learning technique that combines multiple weak learners (usually decision trees) to create a stronger predictive model. The final prediction is a weighted combination of all models, resulting in high accuracy, robustness, and insights into feature importance.

This talk aims to provide mathematical modelers with an understanding of explainable ML in early cancer diagnosis. Our journey bridges the gap between the established hallmarks of cancer and the nascent, yet critical, early signatures. By understanding these precursors, we empower clinicians, researchers, and patients to intervene earlier, potentially altering the course of the disease.