Predicting residual strength of hybrid fibre-reinforced Self-compacting concrete (HFR-SCC) exposed to elevated temperatures using machine learning

Abstract

Hybrid fibre-reinforced Self-compacting concrete (HFR-SCC) offers significant advantages over conventional concrete like increased ductility, crack resistance, and eliminating the need for compaction etc. However, it is very difficult to determine residual strength properties of HFR-SCC after a fire event since it requires rigorous experimental work and resources. Thus, this research presents innovative ways for reliable prediction of compressive strength (cs), flexural strength (fs), and tensile strength (ts) of HFR-SCC using different machine learning (ML) algorithms including gene expression programming (GEP), adaptive neuro-fuzzy inference system (ANFIS), extreme gradient boosting (XGB), AdaBoost, and random forest regression (RFR). The data to be used for this purpose was obtained from internationally published literature having nine inputs including cement, fly ash, temperature, fibre content etc. and three output parameters i.e., cs, ts, and fs. The collected dataset was split into two sets named training and testing sets to be used for training the algorithms and testing their accuracy respectively. The developed predictive models were validated by error metrices including coefficient of determination $\left(R^2\right)$, performance index (PI), and a20-index, etc. The comparison of the algorithms revealed that XGB surpassed its counterparts having testing $R^2$ values equal to 0.998, 0.997, and 0.999 for cs, ts, and fs prediction respectively. Also, the PI values were the lowest for XGB-based predictive model in both phases of training and testing. Thus, Shapely Additive Analysis (SHAP) was performed on the XGB model which revealed that temperature, fibre content, and cement are some of the main contributors to predict the three outputs. The developed predictive models presented in this study can be utilized effectively by the professionals to estimate the residual strength of HFR-SCC.