A Hybrid Deep Learning–Driven Privacy Preservation Framework for High-Availability Big Data Systems

The rapid expansion of big data ecosystems demands privacy-preserving mechanisms that ensure confidentiality without degrading data availability. Existing models—such as k-anonymity, differential privacy, and deep-learning–based anonymization—often face limitations in scalability, utility retention, and adaptive threat response. Building upon prior studies on hierarchical feature learning and availability-centric anonymization, this research proposes a Hybrid Deep Learning–Driven Privacy Preservation Framework (HDL-PPF) that integrates deep feature abstraction, adaptive anonymization, entropy-optimized perturbation, and real-time threat intelligence. Inspired partly by the architectural concepts in SDL-DSA (Azharudheen et al., 2025), this framework enhances privacy strength while improving utility and reducing computational latency. Experimental evaluations using benchmark datasets demonstrate improved privacy index, lower reconstruction risk, stronger anomaly detection accuracy, and higher availability when compared with previous models such as CDBN-RCDO and autoencoder-based mechanisms. The proposed framework provides a scalable security model suitable for cloud, healthcare, IoT, and financial big data systems.