View in EDS

A Comprehensive Review of Bias in AI, ML, and DL Models: Methods, Impacts, and Future Directions.

Saved in:
Bibliographic Details
Title: A Comprehensive Review of Bias in AI, ML, and DL Models: Methods, Impacts, and Future Directions.
Authors: Kumar, Ankur1,2 (AUTHOR) khatriankur007@yahoo.com, Dhanka, Sanjay3,4 (AUTHOR) sanjaykumar506070@gmail.com, Sharma, Abhinav5 (AUTHOR) abhinav.engi@gmail.com, Sharma, Anchal1 (AUTHOR) anchal.15000@gmail.com, Nain, Monika6,7 (AUTHOR) nainmonika551@gmail.com, Kumar, Prashant8 (AUTHOR) prashant.engineering@tmu.ac.in, Gupta, Atma Ram9 (AUTHOR) arguptanitd@gmail.com, Bansal, Jyoti10 (AUTHOR) principalbfcet@babafaridgroup.com, Saxena, Nitin Kumar11 (AUTHOR) nitinsaxena.iitd@gmail.com, Pant, Ruby12 (AUTHOR) pant.ruby12@gmail.com
Source: Archives of Computational Methods in Engineering. May2026, Vol. 33 Issue 4, p5743-5773. 31p.
Subjects: Algorithmic bias, Artificial intelligence & ethics, Artificial intelligence
Abstract: Bias in artificial intelligence (AI), machine learning (ML), and deep learning (DL) models presents a critical challenge to achieving fairness and trustworthiness in high-stakes fields like healthcare, finance, and criminal justice. Documented instances include facial recognition systems failing significantly more often on darker-skinned women and healthcare algorithms systematically underestimating the care needs of Black patients due to flawed data proxies. This study offers a comprehensive review of bias in AI, analyzing its sources, detection methods, and bias mitigation strategies. The authors systematically trace how bias propagates throughout the entire AI lifecycle, from initial data collection to final model deployment. The review then evaluates state-of-the-art mitigation techniques, such as pre-processing (e.g. data re-sampling), in-processing (e.g. adversarial debiasing), and post-processing methods. A recurring theme identified is the fairness-accuracy trade-off, where efforts to improve group fairness by 10–15% often result in a modest 2–5% reduction in overall model accuracy. Through case studies in hiring, healthcare, and predictive policing, this work illustrates the real-world applicability, strengths, and limitations of these approaches. Furthermore, this research examines the evolving legal and ethical landscape, including frameworks like the EU AI Act and the "Right to Explanation" under GDPR, underscoring the necessity for regulatory compliance and interdisciplinary collaboration. Key challenges such as scalable fairness auditing and context-dependent fairness definitions are highlighted. Finally, this study discusses emerging trends like bias-aware federated learning and explainable AI as promising future research directions, providing a roadmap for developing more transparent, inclusive, and equitable AI systems. Highlights: Taxonomy of AI bias: Data, algorithmic, societal, evaluation. Bias mitigation: Pre-, in-, post-processing techniques reviewed. Real-world impacts: Healthcare, finance, justice, employment cases. Legal & ethics: GDPR, EU AI Act, IEEE guidelines analyzed. Future directions: Scalability, federated AI, interdisciplinary solutions. [ABSTRACT FROM AUTHOR]
Copyright of Archives of Computational Methods in Engineering is the property of Springer Nature and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:Bias in artificial intelligence (AI), machine learning (ML), and deep learning (DL) models presents a critical challenge to achieving fairness and trustworthiness in high-stakes fields like healthcare, finance, and criminal justice. Documented instances include facial recognition systems failing significantly more often on darker-skinned women and healthcare algorithms systematically underestimating the care needs of Black patients due to flawed data proxies. This study offers a comprehensive review of bias in AI, analyzing its sources, detection methods, and bias mitigation strategies. The authors systematically trace how bias propagates throughout the entire AI lifecycle, from initial data collection to final model deployment. The review then evaluates state-of-the-art mitigation techniques, such as pre-processing (e.g. data re-sampling), in-processing (e.g. adversarial debiasing), and post-processing methods. A recurring theme identified is the fairness-accuracy trade-off, where efforts to improve group fairness by 10–15% often result in a modest 2–5% reduction in overall model accuracy. Through case studies in hiring, healthcare, and predictive policing, this work illustrates the real-world applicability, strengths, and limitations of these approaches. Furthermore, this research examines the evolving legal and ethical landscape, including frameworks like the EU AI Act and the "Right to Explanation" under GDPR, underscoring the necessity for regulatory compliance and interdisciplinary collaboration. Key challenges such as scalable fairness auditing and context-dependent fairness definitions are highlighted. Finally, this study discusses emerging trends like bias-aware federated learning and explainable AI as promising future research directions, providing a roadmap for developing more transparent, inclusive, and equitable AI systems. Highlights: Taxonomy of AI bias: Data, algorithmic, societal, evaluation. Bias mitigation: Pre-, in-, post-processing techniques reviewed. Real-world impacts: Healthcare, finance, justice, employment cases. Legal & ethics: GDPR, EU AI Act, IEEE guidelines analyzed. Future directions: Scalability, federated AI, interdisciplinary solutions. [ABSTRACT FROM AUTHOR]
ISSN:11343060
DOI:10.1007/s11831-025-10483-6