Investigating the durability and sustainability of soilcrete containing metakaolin using adaptive neuro‒fuzzy inference system

Recent years have witnessed a burgeoning interest in sustainable, eco-friendly, and cost-effective construction materials for civil engineering projects. Soilcrete, an innovative blend of soil and cement, has gained significant acclaim for its versatility and effectiveness. It serves not only as grout for soil stabilization in corrosive environments like landfills and coastal regions but also as a reliable material for constructing structural elements. Understanding the mechanical properties of soilcrete is crucial, yet traditional laboratory tests are prohibitively expensive, time-consuming, and often imprecise. Machine learning (ML) algorithms present a superior alternative, offering efficiency and accuracy. This research focuses on the application of the adaptive neuro-fuzzy inference system (ANFIS) algorithm to predict the uniaxial compressive strength (UCS) of soilcrete. A total of 300 soilcrete specimens, crafted from two types of soil (clay and limestone) and enhanced with metakaolin as a pozzolanic additive, were meticulously prepared and tested. The dataset was divided, with 80% used for training and 20% for testing the model. Eight parameters were identified as key determinants of soilcrete’s UCS: soil type, metakaolin content, superplasticizer content, shrinkage, water-to-binder ratio, binder type, ultrasonic velocity, and density. The analysis demonstrated that the ANFIS algorithm could predict the UCS of soilcrete with remarkable accuracy. By combining laboratory results with ANFIS model predictions, the study identified the optimal conditions for maximizing soilcrete's UCS: 11% metakaolin content, a 0.45 water-to-binder ratio, and 1% superplasticizer content.