Prediction of flexural strength of FRC pavements by soft computing techniques

Purpose: The mechanical characteristics of concrete used in rigid pavements can be improved by using fibre-reinforced concrete. The purpose of the study was to predict the flexural strength of the fibre-reinforced concrete for ten input variables i.e., cement, fine aggregate, coarse aggregate, water, superplasticizer/high range water reducer, glass fibre, polypropylene fibre, steel fibres, length and diameter of fibre and further to perform the sensitivity analysis to determine the most sensitive input variable which affects the flexural strength of the said fibre-reinforced concrete. Design/methodology/approach: The data used in the study was acquired from the published literature to create the soft computing modes. Four soft computing techniques i.e., Artificial neural networks (ANN), Random forests (RF), Random trees RT), and M5P, were applied to predict the flexural strength of fibre-reinforced concrete for rigid pavement using ten significant input variables as stated in the ‘purpose’. The most performing algorithm was determined after evaluating the applied models on the threshold of five statistical indices, i.e., the coefficient of correlation, mean absolute error, root mean square error, relative absolute error, and root relative squared error. The sensitivity analysis for most sensitive input variable was performed with out-performing model, i.e., ANN. Findings: The testing stage findings show that the Artificial neural networks model outperformed other applicable models, having the highest coefficient of correlation (0.9408), the lowest mean absolute error (0.8292), and the lowest root mean squared error (1.1285). Furthermore, the sensitivity analysis was performed using the artificial neural networks model. The results demonstrate that polypropylene fibre-reinforced concrete significantly influences the prediction of the flexural strength of fibre-reinforced concrete. Research limitations/implications: Large datasets may enhance machine learning technique performance. Originality/value: The article's novelty is that the most suitable model amongst the four applied techniques has been identified, which gives far better accuracy in predicting flexural strength.