Estimating the Strength of Stabilized Dispersive Soil with Cement Clinker and Fly Ash

In this study, the potential of four popular artificial intelligence techniques random forest (RF), Gaussian process (GP), M5P tree and artificial neural network (ANN) are assessed for estimating the strength of stabilized dispersive soil with cement clinker and fly ash. GP, M5P and ANN models were providing a good estimate of performance, whereas the RF model outperforms them. For this study, a dataset containing 52 observations obtained from the laboratory experiments. Total data set (52 observations) has been segregated in two different groups. The larger group (36) was used for model development and the smaller group (16) was used for testing the models. Input dataset consists of dispersive soil (%), cement clinker (%), fly ash (%) and curing time (days), whereas unconfined compressive strength (UCS) of soil (MPa) was taken as a target. Sensitivity testing results conclude that the curing time is the most essential factor in estimating the strength of dispersive soil with cement clinker and fly ash for RF-based modelling. The results of this study also suggest that the combined mix of cement clinker and fly ash are used to increase the UCS of dispersive soil than an alone mix.