Machine Learning Prediction of Plasma‐Activated Medium Efficacy to Enable Precision Dosing in Oncology

Cold atmospheric plasma‐activated medium exhibits selective anticancer activity, yet its efficacy is sensitive to controllable parameters, limiting reproducibility. To enable empirical optimization, we curated 749 experimental observations (2013–2025) and developed machine learning models predicting therapeutic outcomes. Using 15 predictors covering plasma generation and cell‐line features, tree‐based ensembles outperformed other algorithms. An optimized Random Forest regressor achieved R ² = 0.7505, while a tuned LightGBM classifier reached 76.42% accuracy (macro‐AUC 0.92). Feature analysis identified treatment time, volume, and usage as primary determinants, with gas type and FBS concentration contributing minimally. These findings support protocol optimization and a standardized “plasma dose,” advancing clinical translation toward precision, cell‐context–aware dosing.