Machine Learning on Systematically Curated Data Reveals Key Determinants of Magnetic Hyperthermia Performance

ABSTRACT Accurate prediction of the specific absorption rate (SAR) of superparamagnetic iron oxide nanoparticles (SPIONs) is critical for optimizing their performance in magnetic hyperthermia applications. This study presents the development of a predictive model for SAR using advanced machine learning techniques and a systematically curated dataset comprising 1850 entries from 84 published studies, capturing 30 predictive features related to SPION properties and experimental parameters. Twelve machine learning algorithms were evaluated and optimized using Bayesian hyperparameter tuning. The CatBoost algorithm emerged as the top‐performing model ( R 2 = 0.98) with the lowest prediction errors. Shapley additive explanation analysis revealed alternating magnetic field amplitude and frequency as the most influential factors determining SAR, followed by SPION concentration and core surface area. Model reliability was confirmed through conformal prediction, providing a prediction interval of ±62 W g −1 . Validation using an independent dataset of SPIONs with varying sizes (7–30 nm) and dopants (Zn, Mn, Mg, Co) demonstrated strong predictive performance for small nanoparticles (≈7 nm), with increased variability for larger particles. These findings demonstrate that advanced machine learning models enable accurate SAR prediction and provide critical insights into nanoparticle design, supporting the systematic optimization of SPIONs for clinical magnetic hyperthermia applications.