Shrink–swell index prediction through deep learning

Abstract Growing application of artificial intelligence in geotechnical engineering has been observed; however, its ability to predict the properties and nonlinear behaviour reactive soil is currently not well considered. Although previous studies provided linear correlations between shrink–swell index Atterberg limits, obtained model accuracy values were found unsatisfactory results. Artificial intelligence, specifically deep learning, potential give improved accuracy. This research employed learning more accurate indices, which explored two scenarios; Scenario 1 used features liquid limit, plastic plasticity index, shrinkage, whilst 2 added input feature, fines percentage passing through a 0.075-mm sieve (%fines). Findings indicated that implementation neural networks resulted increased measurement Scenarios 2. The measured this study suggestively higher have wider variance than most studies. Global sensitivity analyses also conducted investigate influence each feature. These range predicted within data 2, with %fines having highest contribution relevant interaction shrinkage %fines. proposed was around 10–65% preceding models considered study, can then be expeditiously estimate indices.