<p>Power system faults disrupt the stable operation of modern electric grids, and accurate estimation of fault severity supports efficient repair management and the development of mitigation strategies. Conventional approaches focus on fault detection and classification but do not quantify the continuous impact of faults. This study introduces a shift from traditional classification toward regression‑based prediction of fault severity through a novel integration of Support Vector Regression and Histogram‑Based Gradient Boosting. The base models are optimized using Horse Herd Optimization and Henry Gas Solubility Optimization, enabling reduced feature dimensionality and refined hyperparameters. This unified framework captures non‑linear fault dynamics while preserving interpretability. The best hybrid model achieves RMSE of 0.147 and R² of 0.985 in test phase. The framework also isolates key operational variables, such as power load, that directly influence fault duration. This interpretable approach forms a scalable tool for continuous fault monitoring and strengthens grid resilience by enabling proactive maintenance strategies.</p> Graphical abstract <p></p>