Abstract The rapid adoption of machine learning (ML) in cybersecurity has significantly improved threat detection and response capabilities. However, conventional ML-based security systems often require centralized data collection, leading to serious privacy risks such as data leakage, unauthorized access, and regulatory non-compliance. This paper explores privacy-focused machine learning models for cybersecurity, emphasizing techniques that ensure data confidentiality while maintaining detection accuracy. Approaches such as federated learning, differential privacy, homomorphic encryption, and secure multi-party computation are analyzed in the context of intrusion detection, malware classification, and anomaly detection. The study highlights the trade-offs between privacy, performance, and computational overhead, and discusses real-world challenges in deploying privacy-preserving ML systems. The findings suggest that privacy-aware ML frameworks are essential for building trustworthy, scalable, and regulation-compliant cybersecurity solutions.
Radhika et al. (2026) studied this question.