Feature Selection with RVM and Its Application to Prediction Modeling

Abstract

We describe here a method named FSRVM-PLS for model construction using relevance vector machine (RVM). The most compelling feature of FSRVM-PLS is that it's not necessary to estimate parameters in the feature selection phase benefiting from a fully probabilistic framework. After evaluating the effectiveness of FSRVM on a synthetic data set, our method is applied successfully to the prediction of aqueous solubility and permeability.