Explainable Machine Learning for Performance Assessment of Fiber-Nanoparticle Concrete in High-Altitude Cold...

The construction of concrete structures in high-altitude cold regions faces unique challenges, including intense radiation, low atmospheric pressure, frequent freeze-thaw cycles, and salt erosion, which critically impact long-term durability. Accurately predicting compressive strength is vital for optimizing mix designs and ensuring structural safety. While existing studies focus on plains, marine, or hilly environments, high-altitude regions remain understudied. This study proposes a novel hybrid model integrating convolutional neural networks, long short-term memory, and extreme gradient boosting (XGBoost). To the best of our knowledge, this is one of the first studies to specifically address the predictive modeling of concrete durability under the coupled effects of high-altitude environmental factors (low pressure, high radiation, freeze-thaw, and salt erosion). A comprehensive data set of 620 compressive strength samples was curated, with balanced training/testing sets via K-means++ clustering and 5-fold cross-validation. Hyperparameter optimization further refined model performance. The hybrid model achieved <1.5% prediction error, outperforming 12 traditional models. Robustness analysis confirmed stability, while the application of SHAP successfully effectively interpreted the "black-box" model, quantifying the nonlinear contributions of key factors influencing strength, such as water-cement ratio, additives, and curing conditions. This approach provides a reliable tool for predicting compressive strength in extreme environments, guiding mix design for fiber-reinforced concrete and enhancing durability assessments. Future work will expand applications to other harsh climates and validate against field data.