PREDICTION MODELS WITH SOIL INDEX PARAMETERS

Abstract

Reliable estimation of soil compaction parameters, namely Maximum Dry Density (MDD) and Optimum Moisture Content (OMC), is critical for the design and construction of embankments, pavements, and earth dams. While these parameters are traditionally determined through labour-intensive laboratory tests, this study proposes a computational framework to predict them using easily obtainable soil index properties.

The study started by characterizing soil samples in a lab to ascertain their grain size distribution (gravel, sand, and silt content) and Atterberg limits (Liquid Limit, Plastic Limit, and Plasticity Index). A Multiple Linear Regression (MLR) model was developed using these index attributes as input variables. A Genetic Algorithm (GA) was incorporated into the process to optimize the regression coefficients through iterative selection, crossover, and mutation procedures in order to further improve the predictive capability and reduce the Mean Square Error (MSE).

MSE and the Coefficient of Determination (R²), were used to assess the models' performance. The GA-optimized model fits experimental laboratory values more closely than the regular MLR model, according to a comparative analysis. The results show that combining soft computing methods with conventional geotechnical data provides a reliable, quick, and precise substitute for forecasting compaction characteristics, hence eliminating the need for lengthy, repetitive laboratory testing.