A comparative study on the application of artificial intelligence networks versus regression analysis for the prediction of clay plasticity

Abstract

Plasticity is the significant integrated property of clay-water relationship that can be initially associated with the consistency which is an outstanding term used especially in cohesive soils to describe the geotechnical behavior characteristics depending upon the change of water content. In the context of this study, the consistency characteristics of plastic clays are investigated based on the analyses conducted with both applications of regression and artificial intelligence methods. In order to acquire an actual input mesh, a domain-specific dataset has been created with the evaluation of 350 soil investigation reports containing a huge number of consistency tests, including the districts located on the European side of Istanbul, Turkey. The results of the conducted laboratory tests are recorded for very high and high plastic clays which are dominantly situated in the southwest regions of Istanbul. Regression analyses and artificial intelligence techniques have been performed with a frequently used software to query the attainment of the values of plastic limit and plasticity index directly from only liquid limit test results. The main aim to acquire a direct relationship between the liquid limit versus plasticity index value is to reduce the dependency of the consistency limit tests to the operators' experience and also the physical condition of the experimental application environment. As a result of the analyses carried out for this purpose, equations with sufficient reliability, which are applied to obtain the direct plasticity index, were acquired. This condition enables to eliminate the application of the tests of the plastic limit. At the same time, the concurrency of the technique used in obtaining the related equations was questioned comparatively by using regression analysis and artificial intelligence applications. Consequently, discussions are made with well-known studies of literature to validate the applicability of the obtained relationships.