SYSTEMATIC LITERATURE REVIEW: VALIDATION METHODS FOR MACHINE LEARNING MODELS IN IOT CYBERSECURITY

Evaluating Current Validation Practices and Future Directions

Abstract

Machine learning (ML) has emerged as a critical enabler of cybersecurity in Internet of Things (IoT) environments, offering adaptive and scalable mechanisms for detecting evolving threats. However, the reliability and deployment readiness of ML-based cybersecurity models depend heavily on the rigor and contextual relevance of their validation approaches. This study presents a systematic literature review (SLR) examining validation techniques applied in ML-driven IoT cybersecurity research, synthesizing findings from 54 peer-reviewed articles published between 2018 and 2024 across IEEE Xplore, SpringerLink, ScienceDirect, and ACM Digital Library. The review spans various application domains, including intrusion detection, malware classification, threat prediction, and adversarial defense.

The findings reveal a strong reliance on traditional techniques such as k-fold cross-validation and hold-out methods, which fall short in addressing IoT-specific challenges, namely class imbalance, temporal drift, adversarial manipulation, and operational heterogeneity. More robust validation methods, including temporal, cross-dataset, and hybrid strategies, remain underutilized in existing literature. To address these gaps, the study proposes a domain-aligned validation framework that integrates time-aware, robustness-focused, and deployment-oriented evaluation strategies. The review offers a structured taxonomy of validation practices and provides actionable insights for improving the empirical rigor of ML-based cybersecurity systems. Beneficiaries include researchers, IoT developers, cybersecurity practitioners, and policymakers aiming to advance trustworthy and context-resilient ML solutions.