An Automatic Relevance Determination Prior Bayesian Neural Network for Controlled Variable Selection

TL;DR

This paper introduces a Bayesian Neural Network with an Automatic Relevance Determination prior that improves feature importance measurement, enhancing variable selection and prediction accuracy in real-world datasets.

Contribution

It proposes a novel ARD prior-based Bayesian Neural Network for feature importance, demonstrating superior performance over existing methods in variable selection and prediction.

Findings

Significant improvement in variable selection power.

Enhanced predictive performance on real-world data.

Effective feature importance measurement using ARD prior.

Abstract

We present an Automatic Relevance Determination prior Bayesian Neural Network(BNN-ARD) weight l2-norm measure as a feature importance statistic for the model-x knockoff filter. We show on both simulated data and the Norwegian wind farm dataset that the proposed feature importance statistic yields statistically significant improvements relative to similar feature importance measures in both variable selection power and predictive performance on a real world dataset.