Axial compressive strength predictive models for recycled aggregate concrete filled circular steel tube columns using ANN, GEP, and MLR

In recent years, recycled aggregate concrete (RAC) has been used as a suitable solution to solve the problems related to the disposal of construction waste and contribute to sustainable environmental development. However, RAC has defects such as low modulus of elasticity and low compressive bearing capacity, which limits the use of RAC only in non-structural cases. Using composite columns in construction is an excellent way to overcome this limitation. Thus, increasing the use of these columns makes it possible to predict their compressive strength to assist in their design. As part of this study, several artificial intelligence algorithms, including the ANN, the GEP, and the MLR, have been analyzed to predict the compressive strength of recycled aggregate concrete-filled circular steel tubes (RACFCST). A total number of 103 valid experimental data were collected from earlier investigations and the models were trained by 75% of them; the rest were considered for testing. RACFCST's compressive strength was investigated as the target of the models. In conclusion, the proposed models were evaluated by contrasting them to code equations like ACI, Eurocode, AIJ and DL/T to determine whether or not they were valid and whether or not they were predictable. The findings of this research suggested that all three models utilized in this investigation produced accurate predictions when compared to the results of the experiments. As a consequence, the values of R2 in all three models were greater than 0.9, but the ANN model, which had an R2 value of 0.993, demonstrated the highest level of accuracy. While among design code equations, DL/T with R2 equal to 0.957 showed the best performance in predicting the results. •Estimation of compressive strength of recycled aggregate concrete filled circular steel tube columns.•Innovative models and explicit equations for RACFCST compressive strength prediction via soft computing methods.•Existing design codes and equations are presented.•Developed models and existing design codes outcomes were compared.