Classifying and predicting blast risks behind protection systems using machine learning models

Terrorist attacks involving explosions present serious safety risks to the public and essential buildings, highlighting the urgent need for effective protection systems. However, the development and implementation of such systems face significant challenges, including prolonged design processes, costly performance verification tests, and complex simulation models that require high computational demands. These challenges limit the ability to rapidly evaluate and optimize different blast mitigation system configurations under diverse explosion characteristics. This study addresses these limitations by developing and validating a data-driven framework that employs machine learning models to classify and predict transmitted pressure values behind a Bilayer Lightweight Arched Steel (BLASt) mitigation system. Initially, descriptive analyses are conducted to examine various explosion characteristics and system parameters influencing system performance. Two classification models, CHAID and C5.0, are then applied to classify protection levels, achieving high accuracy levels of 96% and 97%, respectively, demonstrating their effectiveness in evaluating system performance. For pressure prediction, three machine learning approaches are utilized: Artificial Neural Networks (ANN), Random Forest (RF), and Genetic Programming (GP). The GP model is further used to generate two mathematical expressions capable of directly predicting transmitted pressure values. All developed models are validated through regression analyses and compared against actual pressure values, showing strong agreement and high prediction accuracy, with R 2 values ranging from 0.91 to 0.95. Furthermore, interpretability analyses reveal that scaled distance is the dominant factor influencing transmitted pressure values, contributing approximately 72%, while system parameters account for around 28%. Among the models, GP provides accurate and practical predictive expressions with minimal error. Overall, the findings demonstrate that machine learning models offer an efficient and reliable approach for classifying and predicting blast pressures, supporting improved design and implementation of protection systems under sudden-threatening scenarios.