notesum.ai

Published at November 27

A Machine Learning-based Framework towards Assessment of Decision-Makers' Biases

cs.LG

Released Date: November 27, 2024

Authors: Wanxue Dong¹, Maria De-arteaga², Maytal Saar-Tsechansky²

Aff.: ¹Department of Decisions, Operations and Technology, The Chinese University of Hong Kong Business School; ²The Department of Information, Risk, and Operations Management, McCombs School of Business, The University of Texas at Austin

Arxiv: http://arxiv.org/abs/2411.18122v1

		20% Positive Label Prevalence			30% Positive Label Prevalence
Dataset	GS Per	MDBA	MDBA-	MDBA	MDBA	MDBA-	MDBA
	worker		Naive	IMPROV.		Naive	IMPROV.
Adult	100	0.082	0.099	17.1%	0.071	0.076	7.4%
	200	0.073	0.08	9.1%	0.043	0.05	14.5%*
	300	0.076	0.079	2.9%	0.044	0.052	15.4%
	400	0.066	0.066	0.3%	0.036	0.043	14.6%
Credit	100	0.129	0.184	29.8%***	0.097	0.121	19.4%*
	200	0.099	0.16	38.2%***	0.077	0.105	26.7%**
	300	0.095	0.155	38.6%***	0.074	0.103	28%***
	400	0.089	0.161	44.5%***	0.073	0.101	27.7%***
Employee	100	0.082	0.109	24.7%**	0.072	0.093	22.4%*
	200	0.057	0.076	25.5%***	0.06	0.08	25.1%*
	300	0.045	0.069	34.8%***	0.054	0.077	30.8%**
	400	0.046	0.064	28.1%**	0.049	0.07	30.3%**
Readmission	100	0.081	0.103	21.2%	0.078	0.083	6.3%
	200	0.059	0.089	34%***	0.04	0.042	3.7%
	300	0.052	0.07	26.1%*	0.035	0.036	1.6%
	400	0.05	0.061	19.4%	0.029	0.03	4.2%