Utilizing Artificial Neural Network and Multiple Linear Regression to Model the Compressive Strength of Recycled Geopolymer Concrete

Based on the heterogeneity of concrete constituents as well as variability in compressive strength over many magnitudes for various types of concrete, predictive methods for evaluating the compressive strength have now been given considerable attention. As a result, this research compares the performance of the Artificial Neural Network, ANN, in forecasting the compressive strength of geopolymer recycled concrete (GPRC) based on selected pozzolans (Coal Fly Ash (CFA) and Rice Husk Ash (RHA)) at ages 7, 28, and 56 days to the traditional Multiple Linear Regression, MLR. The compressive strength of GPRC-based CFA and RHA was determined using 65 concrete samples from eight different mixtures. The developed models were based on the experimental results, which used varying material quantities. The ANN and MLR models were built with eight input variables: Ordinary Portland cement (OPC), RHA, CFA, Crushed granite (CG), Cupola Furnace Slag (CFS), Alkaline Solution (AS), Water-Binder Ratio (WB), and Concrete Age (CA), with compressive strength being the only predicted variable. Using MATLAB (R) code, approximately 75% and 25% of the input data were used for training and testing to develop an ANN model for predicting compressive strength, f(cu). For ANN and MLR, the input data were trained and tested using the feedforward back-proportion and backward elimination approaches, respectively. Based on satisfactory performance in terms of means square error MSE, the most likely model architecture containing eight input layers, thirteen hidden layers, and one output layer neurons was chosen after several trials. According to the MLR results, only three input variables, CFA. CG, and CA, are statistically significant with p-values less than 0.05. R-2 = 0.9972, MSE = 0.4177, RMSE = 1.8201, for ANN and R-2 = 0.7410, MSE = 66.6308, RMSE = 290.4370, for MLR. The predicted results demonstrate the proposed model's dependability and computational forecasting capability. The findings of the study have the potential to help a wide range of construction industry in predicting the concrete properties and managing scarce resources.