Shimobe, SatoruKarakan, EyyubSezer, Alper2024-08-252024-08-2520230941-06431433-3058https://doi.org/10.1007/s00521-023-08292-8https://hdl.handle.net/11454/100099Prediction of the ultimate settlement is vital for the assessment of the service life of a structure, particularly when it is underlain by fine-grained soils. As known, this value is a function of the compression index (C-c) of soils, which can simply be found by performance of oedometer tests. For this purpose, more than 2000 test results from past studies were compiled to constitute a database. Then, multiple linear regression analyses were employed to predict the C-c parameter by use of toughness limit (TL) and a function of this parameter, namely the soil state index (SSI). It was noticed that SSI was a better predictor of C-c, in comparison with TL. Prediction ability of many equations from literature was questioned, and it was concluded that these equations were good predictors of their own data. Moving to a generalized behavior, data show a more scattered structure, which needs more sophisticated methods using above-mentioned parameters as inputs. In this regard, artificial neural networks were employed to estimate the C-c by use of single input parameters: TL or SSI. Additionally, a combination of Atterberg limits was also instructed as inputs for prediction of C-c. A comparative analysis of the effects of learning algorithm, input data, and number of neurons in hidden layer was given. It was concluded that the TL and SSI are reasonable predictors of compression index.en10.1007/s00521-023-08292-8info:eu-repo/semantics/closedAccessCompression indexToughness limitArtificial neural networksRegression analysisGeotechnical PropertiesHydraulic ConductivityEmpirical CorrelationsTrial EmbankmentShear-StrengthSoft ClayConsolidationBehaviorPerformanceSettlementsArticleWOS:0009450602000012-s2.0-85148600918Q1Q2