Assessing Compressive Strength of Concrete with Extreme Learning Machine

Abstract

Manual estimation of compressive strength of concrete (CSC) is time consuming and expensive. Soft computing techniques are found better to statistical methods applied to this problem. However, sophisticated prediction models are still lacking and need to be explored. Extreme learning machine (ELM) is a faster and better learning method for artificial neural networks (ANNs) with solitary hidden layer and has enhanced generalization capacity. This article presents an ELM-based forecast for efficient prediction of CSC. A publicly available dataset from UCI repository is used to develop and access the performance of the model. The prediction accuracy of ELM is compared with few machine learning methods such as back propagation neural network (BPNN), support vector machine (SVM), auto-regressive integrated moving average (ARIMA), and least squared estimation (LSE). A comparative study for the prediction of CSC at the curing ages of 28, 56, and 91 days has been carried out using all models. The experimental findings from ELM-based forecasting demonstrate its ability in predicting CSC in a robust manner. On an average, it achieves lowest MAPE of 0.048024, ARV of 0.052872, U of Theil’s statistics (UT) of 0.038772, NMSE of 0.058522, and standard deviation (SD) of 0.256267. Comparative analysis of simulation results and statistical significance test suggests the superiority of ELM-based CSC prediction.