Data Poisoning Attacks in the Training Phase of Machine Learning Models: A Review

Srivastava, Mugdha and Kaushik, Abishek and Loughran, Roisin and McDaid, Kevin (2024) Data Poisoning Attacks in the Training Phase of Machine Learning Models: A Review. In: Proceedings of The 32nd Irish Conference on Artificial Intelligence and Cognitive Science, December 9-10, 2024, Dublin, Republic of Ireland.

PDF Download (1MB)

Abstract

Data Poisoning Attacks (DPAs) can severely impact the performance of Machine Learning (ML) models by manipulating training datasets to introduce errors or biases. The integrity of ML models is crucial for user safety and trust, especially as these models increasingly influence key decision-making processes in safety-critical sectors like finance, healthcare, and law enforcement. As ML technology advances, so do the vulnerabilities of these systems, making the reliability of training data vital for ensuring accurate and dependable model outcomes. This review examines the growing threat of DPAs on ML systems at the training stage, categorizing these attacks into label manipulation, data injection, feature space manipulation, and relationship manipulation. By exploring multiple types of attacks and providing relevant examples, this analysis aims to raise awareness about the significant risks posed by compromised data, which can lead to widespread mistrust in ML systems and cause considerable harm, including financial losses, legal liabilities, and even threats to human lives.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	Data poisoning; artificial intelligence; machine learning; deep learning; cybersecurity; adversarial attacks.
Subjects:	Computer Science
Research Centres:	Regulated Software Research Centre
Depositing User:	Sean McGreal
Date Deposited:	16 Dec 2025 15:26
Last Modified:	16 Dec 2025 15:26
License:	Creative Commons: Attribution-Noncommercial-Share Alike 4.0
URI:	https://eprints.dkit.ie/id/eprint/980

Actions (login required)

View Item

Downloads