Underestimation Bias and Underfitting in Machine Learning

TL;DR

This paper investigates how classification algorithms can introduce or worsen bias, especially underestimation bias linked to regularization, highlighting the need to understand algorithmic bias sources beyond data.

Contribution

It explores the factors contributing to bias in classification algorithms, emphasizing the role of regularization in underestimation bias, an area less studied in existing research.

Findings

Bias can be amplified by regularization techniques.

Underestimation bias is connected to overfitting measures.

Initial research highlights the importance of algorithmic bias sources.

Abstract

Often, what is termed algorithmic bias in machine learning will be due to historic bias in the training data. But sometimes the bias may be introduced (or at least exacerbated) by the algorithm itself. The ways in which algorithms can actually accentuate bias has not received a lot of attention with researchers focusing directly on methods to eliminate bias - no matter the source. In this paper we report on initial research to understand the factors that contribute to bias in classification algorithms. We believe this is important because underestimation bias is inextricably tied to regularization, i.e. measures to address overfitting can accentuate bias.