Disentangled Deep Smoothed Bootstrap for Fair Imbalanced Regression

TL;DR

This paper presents a novel approach combining disentangled VAEs and smoothed bootstrap in latent space to improve imbalanced regression on tabular data, addressing a gap in existing methods mainly focused on classification.

Contribution

It introduces a new data generation technique using disentangled VAEs and smoothed bootstrap for imbalanced regression, enhancing learning on tabular data.

Findings

Outperforms existing methods on benchmark IR datasets

Effective in modeling data distributions with imbalance

Improves regression accuracy on imbalanced data

Abstract

Imbalanced distribution learning is a common and significant challenge in predictive modeling, often reducing the performance of standard algorithms. Although various approaches address this issue, most are tailored to classification problems, with a limited focus on regression. This paper introduces a novel method to improve learning on tabular data within the Imbalanced Regression (IR) framework, which is a critical problem. We propose using Variational Autoencoders (VAEs) to model and define a latent representation of data distributions. However, VAEs can be inefficient with imbalanced data like other standard approaches. To address this, we develop an innovative data generation method that combines a disentangled VAE with a Smoothed Bootstrap applied in the latent space. We evaluate the efficiency of this method through numerical comparisons with competitors on benchmark datasets for IR.