Distributed, Fault Tolerant and Secure Time Allocation Algorithms for Scalable Communication in the Internet of Things

Abstract

The Internet of Things (IoT) enables large-scale networks of resource-constrained devices that continuously generate and exchange data across critical domains such as healthcare, industry, and smart infrastructure. Ensuring scalable communication in such networks requires the joint optimization of resource efficiency, intrusion detection, privacy preservation, and routing reliability. This thesis addresses these challenges by proposing a unified intelligent framework for distributed, secure, and fault-tolerant communication in IoT, with the Internet of Medical Things (IoMT) adopted as the primary validation domain. The first contribution introduces a Time Allocation strat egy based on dynamic Temporal Aggregation, in which the aggregation window Tagg is adaptively selected to transform raw data streams into compact feature vectors before transmission. This strategy significantly reduces computational cost, achiev ing training-time reductions of up to 97.2% and inference-time reductions of up to 97.5% on CICIoMT-2024, with comparable improvements on NF-UNSW-NB15-v2 and WUSTL-EHMS-2020, thereby enhancing scalability and contributing to lower energy consumption in resource-constrained IoT environments. The second contribution pro poses FTL-HLSTM, a Federated Transfer Learning architecture based on Hierarchical Long Short-Term Memory networks for privacy-preserving intrusion detection. The framework addresses the Non-IID nature of distributed IoT data by distinguishing between globally shared and locally specific attack patterns, enabling collaborative learning without exchanging raw data. The model achieved 100.0% binary detec tion accuracy on the evaluated NF-UNSW-NB15-v2 split and 99.74% accuracy on CICIoMT-2024, while reducing training time compared with standard LSTM models. The third contribution develops a proactive fault-tolerant routing mechanism based on Multi-Criteria Decision Analysis using TOPSIS and AHP. The proposed approach computes a dynamic Trust Score for each node according to safety, energy, latency, and packet-loss criteria. Unlike reactive routing protocols, it excludes non-acceptable nodes from routing decisions and adaptively distributes traffic by assigning 70% to optimal nodes and 30% to acceptable nodes, thereby improving routing reliability un der node failures and insider attacks. Finally, these components are integrated into a unified cyber-physical defense system in which intrusion detection results directly sup port fault-tolerant routing decisions. This closed-loop architecture enables autonomous self-protection and self-healing in IoT networks by linking detection intelligence with adaptive network control.