Estimation of Soil Compression Coefficient Using Artificial Neural Network and Multiple Regressions

ABSTRACT: Measurement of some significant properties of soil might be difficult, costly and time-consuming. Thus, estimation of these characteristics using conveniently measurable soil properties may be useful. In this research, it is attempted to evaluate and examine the artificial neural network technique and multiple regression in order to measure the soil compression coefficient using conveniently measurable soil properties. A total of 100 soil samples were taken randomly from various areas of Ahwaz and the percentage of clay, silt, sand, wet bulk density, dry bulk density, friction coefficient, viscosity coefficient, plastic limit were determined as conveniently measurable properties (dependent variable) and the compression coefficient as costly measured properties ( independent variable).In order to form Annual Neural Network, instructional algorithm Mark- Laurinburg and Perseptron stricter was use and the stepwise method was used in order to make regression transfer functions. The compression coefficients of soils were mean 0.16, at least 0.11 and at most 0.25 and depended on clay soil class. The results showed that the compression coefficient r=0.63 and MSE=0.006 was determined by neural network method. In the regression method, it was measured as r=0.47 and MSE=0.002. By comparing the values of correlation coefficient and error square mean by two using methods, it was revealed that artificial neural network has the least error and the most accuracy. Therefore in the study area the practice of this method is recommended for estimating the compression coefficient.