Performance prediction modeling of standard penetration blow count of clayey soils by two non-linear tools

Yesiloglu-Gultekin N.

ARABIAN JOURNAL OF GEOSCIENCES, vol.14, no.3, 2021 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 14 Issue: 3
  • Publication Date: 2021
  • Doi Number: 10.1007/s12517-021-06649-8
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aquatic Science & Fisheries Abstracts (ASFA), Geobase, INSPEC
  • Keywords: Standard penetration test, ANN, ANFIS, Clayey soil, Activity, Consistency index
  • Hacettepe University Affiliated: Yes


The standard penetration test (SPT) is one of the most common site investigation tools used in mining, geological, civil, and petroleum engineering applications because of its low cost, extensive past database, and advantages such as easy procedure and equipment simplicity. The data employed in this study were obtained from different boreholes of a sewerage project in Mersin in Turkey. A total of 202 data points from 34 boreholes were assessed in this study. A total of 202 corrected SPT blow counts (SPT-N-60) and corresponding parameters including the fine-grained percentage (%), water content (W%), activity, and consistency index (I-c) of the clayey soil for developing the models. Artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) models were constructed to predict SPT-N-60. Three different models with two inputs and an output were constructed. These modeling results were used to determine the most successful model, prediction tool, and dataset to use with statistical indices. The Performance Index employed coefficient of determination (R-2), values account for (VAF), and root mean square error (RMSE) to select the best prediction tool and the model. In addition to these indices, degree of consistency (C-d) was used for an accurate and consistent dataset. Consequently, ANN was selected as the most successful prediction tool, Model 1 with activity and water content was selected as the most successful model, and Model 1 SET-3 was selected as the most consistent and reliable dataset.