Praedico – Salvos: An Ensemble ML framework for predicting survivability of thyroid cancer patients

Abstract

This paper introduces Praedico-Salvos, a novel machine-learning framework for predicting the survival of thyroid cancer patients. Praedico-Salvos offers a significant advancement over existing methods by predicting survival in four distinct time ranges, rather than a simple binary outcome. This fine-grained prognosis empowers oncologists to tailor treatment plans by considering factors like pain tolerance, financial limitations and predicted survival probability. The model leverages data from the well-established Surveillance, Epidemiology, and End Results (SEER) program, addressing the critical need for more nuanced prognoses in thyroid cancer treatment. Praedico-Salvos achieves a higher accuracy of 88% compared to previous models due to its unique capabilities: (a) handling missing data without imputation, (b) transcending binary classification limitations, and (c) categorizing survival into four distinct time bins. Future advancements could incorporate regression within these bins, further refining predictions for the month.