Backpropagation Neural Network – Based Prediction of Crack Depth in Fly-Ash Geopolymer Concrete with Akaike and Bayesian Information Criterion

Concrete have been a major contributor in the environmental pollution particularly its the carbon footprint during its production. In connection with the sustainable development goals (SDGs), the use of alternative materials with lower emissions and footprint could alleviate and help in the expanding problem of global warming. This study evaluated the potential of fly-ash (FA) geopolymer concrete (GC) and create a machine learning based model using backpropagation neural network (BP-NN) for predicting crack depth. The results of the modelling exhibited that the governing configuration of the crack forecast model is 2-5-1 (input-hidden-output) (I-H-O) with Mean Absolute Percentage Error (MAPE) of 9.71%. Additional layer of evaluation was implemented in the identification of the governing model which includes the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). Lowest AIC and BIC values were observed in the governing model structure which suggested that it is the best among the features evaluated. Moreover, the variable contribution (VC) to the crack depth model development were assessed using Garson's algorithm (GA) and the results showed that the pulse wave travel time is the parameter with the highest contribution to the crack depth model development.