Parametric evaluation and prediction of design parameters of geofoam using artificial neural network and extreme gradient boosting models

Expanded polystyrene (EPS) geofoam is increasingly used in the construction industry as a lightweight fill material. The selection of an appropriate grade of geofoam in such cases is dictated by their elastic modulus and compressive strength. However, a comprehensive parametric study on the influencing factors of compression behaviour, which is extremely critical for the design of geofoam, is rarely reported. In the present study, the effect of nominal density, apparent density, strain rate, geometry, and ambient temperature on the elastic modulus, permissible compressive stress, yield stress and compressive strength of geofoam is investigated by conducting a series of uniaxial compression tests. The variation of compressive responses due to each influencing factor is evaluated using scatter matrices and statistical bar plots. Furthermore, results reported in the literature were collated, to develop machine learning based generalised prediction models using Artificial Neural Network and Extreme Gradient Boosting algorithms. The XGBoost models demonstrated superior performance compared to the ANN models, achieving accuracies surpassing 87%. The correlation heat map of the results indicates that the apparent density, size, and ambient temperature control the compressive response of geofoam, while model-dependent feature analysis quantified the relative importance of these parameters. For conservative design and quality assurance, testing a 50 mm geofoam cube at a strain rate of 1% per minute, at the maximum ambient temperature of the construction site is recommended. This study enables the design engineers in the selection of the appropriate grade of geofoam and the associated project cost estimation.