Use of the Relevance Vector Machine for Prediction of an Overconsolidation Ratio

This article uses the relevance vector machine (RVM) for the prediction of the overconsolidation ratio (OCR) of fine-grained soils based on piezocone penetration test data. RVM provides an empirical Bayes method of function approximation by kernel basis expansion. It uses the corrected cone resistance (q(t)), vertical total stress (sigma(v)), hydrostatic pore pressure (u(0)), pore pressure at the cone tip (u(1)), and the pore pressure just above the cone base (u(2)) as input parameters. An equation has also been developed for the determination of OCR. The developed RVM model gives the variance of the predicted data. Sensitivity analysis has been conducted for determining the influence of each input parameter. The results are also compared with some of the existing interpretation methods. Comparisons indicate that the developed RVM model performs better than the existing interpretation methods for predicting OCR. (C) 2013 American Society of Civil Engineers.