Energy‐Aware Plasma Degradation of Antibiotics Without Externally Dosed Oxidants Enabled by Interpretable Machine Learning

ABSTRACT

Reducing electricity demand without externally dosed oxidants is essential for resource‐efficient plasma water treatment. We developed an energy‐aware framework integrating machine‐learning prediction, SHAP interpretation, and multi‐objective optimization. Using 716 records from 24 studies, nine algorithms were evaluated for degradation and energy efficiency. CatBoost performed best, achieving test‐set R ² values of 0.9089 and 0.9407, respectively. SHAP identified treatment time and voltage as the dominant predictors of degradation and energy efficiency, revealing their trade‐off. For trimethoprim in deionized water, CatBoost models coupled with NSGA‐II identified energy‐oriented, balanced, and high‐degradation operating windows. Three Pareto conditions were experimentally validated, with errors below 5% and 9%, respectively. The framework avoids external oxidants, although mineralization, toxicity, off‐gas emissions, and life‐cycle impacts require assessment in future studies.