Machine learning-based prediction of thermal conductivity in foamed concrete: Influence of nano-microsilica compounds and air content using XGBoost

The thermal conductivity (k) of foamed concrete is a key factor in building energy efficiency, especially for ceiling applications where effective thermal insulation is required. This study employs a machine learning approach to predict the thermal conductivity of foamed concrete using the eXtreme Gradient Boosting (XGBoost) algorithm. An initial dataset of 28 experimental samples was expanded to 200 samples via a SMOTE-like interpolation technique. The input variables included Temperature (T), Cement content (C), Air content (Air), Latex percentage (Latex), nano-microsilica powder (NMP), and nano-microsilica compound (NMC). The XGBoost model’s hyperparameters were optimized through grid search, resulting in improved predictive accuracy, with R2 and MSE values of 0.907 and 0.0042 W/m·K, respectively. Sensitivity analysis indicated that air content had the highest influence (54 %) on thermal conductivity, followed by latex content (18 %), NMC (11 %), NMP (9 %), and cement content (8 %). The prominent effect of air content is consistent with fundamental heat transfer principles, as air voids serve as thermal barriers within the concrete matrix. These results demonstrate the effectiveness of an optimized XGBoost model for predicting the thermal properties of foamed concrete. A comparative case study further revealed that implementing the optimized foamed concrete mixture in a typical five-story building in a cold climate can reduce annual heating energy consumption by approximately 15.5 %. This significant energy saving highlights the practical impact of mixture optimization on building performance and sustainability.