Advanced Machine Learning Techniques for Predicting the California Bearing Ratio of Soils Stabilized with Cement Kiln Dust

چکیده مقاله

The California Bearing Ratio (CBR) is an essential test used to design pavement thickness. In general, CBR is inherently expensive, time-consuming, and labor-intensive. As an alternative solution, recently, various machine learning models have been developed to estimate the CBR of stabilized soils. However, predicting the soaked CBR value of cement kiln dust (CKD)-stabilized soils has yet to be investigated. To address this, this study employs various machine learning models to predict the soaked CBR value of cement kiln dust (CKD)-stabilized soils. After a comprehensive database from various sources had been prepared, ten machine learning models were developed. The performance of developed models was compared using a variety of error criteria. The results showed that Gradient Boosting has a superior ability in predicting CBR of CKD-stabilized soils, while the SVM model exhibited the poorest performance. To have a better insight, two further analyses were conducted based on the best-performing prediction model. The results suggested that the CBR of untreated soil, plasticity index, CKD content, and curing time have the highest relative influence on CBR values of CKD-stabilized soil. Further, increasing the CBR of untreated soil, CKD content, and curing time leads to an increase in the stabilized soil’s CBR. In contrast, increasing the plasticity index and liquid limit reduces the CBR.