Datastream anonymization for machine learning applications: a systematic review

Abstract

This study presents a systematic literature review (SLR) focused on recent advances in data stream anonymization and sanitization using machine learning techniques. This review aims to identify key trends, methodologies, and research gaps in the area. A comprehensive search across Scopus and Web of Science yielded 118 articles from the past five years. Findings indicate a growing interest in privacy-preserving data processing, with Federated Learning as the leading approach. Key challenges include scalability limitations, privacy risks from potential data leakage, and communication overhead. The review also highlights a significant lack of standardization in datasets and evaluation metrics, with over 145 distinct datasets primarily from image-based repositories. This study underscores the necessity for standardized benchmarks and evaluation protocols to improve comparability in future research and suggests expanding keyword strategies to capture a wider range of related methodologies.