Smart healthcare IoT systems are vulnerable to cyber threats as they deal with sensitive patient information. Problems such as privacy, scalability, and delayed response to threats in distributed healthcare environments challenge centralized security approaches. To mitigate the security challenges of cloud-edge healthcare IoT systems, this paper presents FL-EZTF, a privacy-preserving, Federated Deep Learning and Enhanced Zero Trust Framework. The framework combines federated learning, Enhanced Zero Trust Architecture (E-ZTA), and Secure Access Service Edge (SASE). In this framework, lightweight deep learning models are developed locally at hospitals and various edge nodes without the need to transfer sensitive medical data. In place of raw data, model updates are sent conveniently through a trustaware federated learning process. Simultaneously, E-ZTA performs continuous authentication, micro-segmentation, and access control to rapidly contain threats. The framework is assessed using CIC-IoT-2023, IoT-23, and WESAD datasets. The experimental results show improved accuracy in detection, lower rates of false positives, a significant reduction in the latency of decisions, and enhanced containment as compared to centralized and traditional federated learning.

With the accelerated proliferation of cloud services and web-based applications, exposure to sophisticated cy-ber threats like zero-day vulnerabilities, advanced persistent threats, and application-layer attacks, has sharply increased. Conventional intrusion detection systems, along with cryptographic security mechanisms, often do not fulfill the require-ments of adaptive detection, privacy preservation, and variety of scalability within distributed systems. To alleviate these problems, this paper suggests a cross-layer adaptive security model, which includes, in the secure cloud and web applica-tions, machine learning-based anomaly detection complemented with advanced cryptographic security. The model com-bines, in this context, lightweight local anomaly detection, federated learning, selective privacy, and a deep reinforcement learning-based threat detection and security orchestration. Through federated learning, active participation in the learning process is assured, while the selective privacy mechanism preserves the model parameters. The deep reinforcement learn-ing agent adjusts the interaction, aggregation, and privacy settings according to demands of the adaptive system and the environment. The assessment of the model is realized through the CSE-CIC-IDS2018, CIC-IDS2017, and the CSIC 2010 HTTP benchmark datasets to verify the model in detection, generalization, and operational effectiveness. The results of the performed tests reflect an accuracy of 97.9%, a F1 score of 97.5%, a false positive rate of 1.8%, and a detection latency of 36 ms, surpassing performance of conventional federated and centralized state-of-the-art models. Cross-dataset testing validates the model effectiveness in the presence of highly variable traffic. The results suggest that adaptive orchestration, federated learning, and selective privacy preservation, when combined, substantially boost intrusion detection, decrease communication overhead, and ensure privacy preservation. Therefore, this framework is a scalable and robust approach to intrusion detection within contemporary cloud and web settings.