Ethical AI with Balancing Bias Mitigation and Fairness in Machine Learning Models

The rapid integration of Artificial Intelligence (AI) into critical domains such as healthcare, finance, and criminal justice has raised significant ethical concerns, particularly around bias and fairness in machine learning models. Despite their potential for improving decision-making processes, th...

Full description

Saved in:
Bibliographic Details
Main Authors: Nathim, Khalida Walid, Hameed, Nada Abdulkareem, Salih, Saja Abdulfattah, Taher, Nada Adnan, Salman, Hayder Mahmood, Chornomordenko, Dmytro
Format: Conference Proceeding
Language:English
Subjects:
Adaptation models
algorithmic bias
and systematic review
artificial intelligence (ai)
bias mitigation
Collaboration
Context modeling
Costs
equity
ethical ai
Ethics
fairness
fairness metrics
Finance
Machine learning
Machine learning algorithms
Medical services
model performance
Prevention and mitigation
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The rapid integration of Artificial Intelligence (AI) into critical domains such as healthcare, finance, and criminal justice has raised significant ethical concerns, particularly around bias and fairness in machine learning models. Despite their potential for improving decision-making processes, these models can perpetuate or even exacerbate existing societal biases. This study aims to investigate approaches to bias mitigation in AI systems, focusing on balancing fairness and performance. A systematic review of 150 research articles published between 2018 and 2023 was conducted, along with experiments on 25 benchmark datasets to evaluate various machine learning algorithms and bias mitigation techniques. Results showed a 23% reduction in bias and an average 17% improvement in nine fairness metrics during model training, though at the cost of up to 9% in overall accuracy. The study highlights the trade-offs between fairness and performance, suggesting that creating AI systems that are both fair and effective remains an ongoing challenge. The findings underscore the need for adaptive frameworks that address bias without significantly compromising model performance. Future research should explore domain-specific adaptations and scalable solutions for integrating fairness throughout the AI development process to ensure more equitable outcomes.
ISSN:2305-7254
2305-7254
2343-0737
DOI:10.23919/FRUCT64283.2024.10749873