PREDICTING SOAKED CALIFORNIA BEARING RATIO OF A-2 LATERITIC SOIL USING SUPPORT VECTOR MACHINE AND RANDOM FOREST IN DELTA STATE

Abstract

It normally takes four days or more in the laboratory to determine the soaked California bearing ratio (CBR) of lateritic soil samples. Furthermore, it is virtually impossible to conduct a large number of tests for a significant project within a short
time frame. Alternative approaches, like forecasting models for soaked California bearing ratio, may therefore be used. The soaked California bearing ratio of A-2 lateritic soil was estimated using the Support Vector Machine (SVM) and Random
Forest in western Niger Delta. A total of 52 dataset samples, comprising the plastic limit, liquid limit, plasticity index, percent of sand that passed through a 200-mesh sieve, moisture content, maximum dry density, and soaked CBR from a published
source and laboratory results from a field investigation, were collected. Wakaito Environment for Knowledge (WEKA) 3.9.5 software was used to investigate the potential of SVMs and random forests to predict the soaked CBR of A-2 lateritic soil. It was found that support vector machine models outperformed random forest models in terms of estimating the soaked California bearing ratio of A-2 lateritic soil. The points on a scatter plot showing outputs from the training, cross-validation and percentage split 65% processes are very close to equality line.