Onyelowe, Kennedy C.Ebid, Ahmed M.Onyia, Michael EAmanamba, Ezenwa C.2023-06-282023-06-282022Onyelowe, K. C., Ebid, A. M., Onyia, M. E., & Amanamba, E. C. (2022). Estimating the swelling potential of non-carbon–based binder (NCBB)-treated clayey soil for sustainable green subgrade using AI (GP, ANN and EPR) techniques. International Journal of Low-Carbon Technologies, 17, 807-815.https://doi.org/10.1093/ijlct/ctac0581748-1325https://nru.uncst.go.ug/handle/123456789/9023A zero carbon footprint stabilization approach has been adopted in this research to improve the swelling potential (SP) of clayey soils for a greener construction approach. Construction activities like earthworks during the cement stabilization of unstable soils utilized as reconstituted subgrade materials is responsible for the emission of unhealthy amount of carbon oxides into the atmosphere contributing to ozone layer depletion and eventual global warming. This has been substituted by using eco-friendly cementing materials, quicklime activated rice husk ash (QARHA), formulated in this research work. The SP of clayey soil treated with QARHA has been predicted using the learning abilities of genetic programming (GP), artificial neural network (ANN) and the evolutionary polynomial regression (EPR). This was aimed at reducing the over dependence on repeated laboratory visits and experiments prior to infrastructure (pavement) designs, construction and future monitoring of the performance of the facility. Multiple data were collected from multiple experiments based on the tested emergent material (QARHA) treatment proportions used in this work. The data were subjected to statistical analysis and predictive model exercises. At the end, the predicted models were validated on the basis of performance and accuracy. The performance indices showed that EPR and GP with R2 of 0.997 outclassed ANN with R2 of 0.994, but EPR outclassed the two, GP and ANN with a minimal error of 6.1%. The performances of GP, ANN and EPR were compared with a previously conducted model, which utilized the learning techniques of the adaptive neuro-fuzzy interface system (ANFIS) and it was observed that EPR and GP performed better than ANFIS but ANN performed at par with it. Generally, the predictive models can predict the SP of subgrade soil treated with QARHA, a non-carbon–based binder with accuracy above 90%, which is a very good outcome.enAI (GP, ANN and EPR) techniquesSoilNon-carbon–based binderArticle