Machine learning approach for predicting compressive strength in foam concrete under varying mix designs and curing periods

Abstract

Efforts to reduce the weight of buildings and structures, counteract the seismic threat to human life, and cut down on construction expenses are widespread. A strategy employed to address these challenges involves the adoption of foam concrete. Unlike traditional concrete, foam concrete maintains the standard concrete composition but excludes coarse aggregates, substituting them with a foam agent. This alteration serves a dual purpose: diminishing the concrete’s overall weight, thereby achieving a lower density than regular concrete, and creating voids within the material due to the foam agent, resulting in excellent thermal conductivity. This article delves into the presentation of statistical models utilizing three different methods—linear (LR), non-linear (NLR), and artificial neural network (ANN)—to predict the compressive strength of foam concrete. These models are formulated based on a dataset of 97 sets of experimental data sourced from prior research endeavors. A comparative evaluation of the outcomes is subsequently conducted, leveraging statistical benchmarks like the coefficient of determination (R2), root mean square error (RMSE), and mean absolute error (MAE), with the aim of identifying the most proficient model. The results underscore the remarkable effectiveness of the ANN model. This is evident in the ANN model’s R2 value, which surpasses that of the LR model by 36% and the non-linear model by 22%. Furthermore, the ANN model demonstrates significantly lower MAE and RMSE values compared to both the LR and NLR models.