Enhancing Reliability of Internet of Things Systems Through Machine Learning-Based and Data Fusion Techniques

The rapid proliferation of Internet of Things (IoT) devices has led to an exponential increase in connected devices across various sectors. Each IoT network typically operates independently, focusing on specific applications and utilizing data fusion techniques to combine sensor-generated data within its network to make a decision. Alongside the growth and development of IoT, vision of sustainable ubiquitous environments, such as smart cities, smart agriculture and health, smart societies etc. is also growing rapidly. Thus, as the number of IoT devices continues to grow and their applications and vision diversify, there arises a pressing need to create a cohesive and intelligent ecosystem of intelligent independent IoT networks. However, IoT systems face significant challenges, particularly in ensuring data reliability and continuity in scenarios of data loss or failure. Such failures, arising from device malfunctions, network disruptions, or environmental factors, result in incomplete or unreliable datasets, which can severely impact decision-making processes and system functionality. Therefore, this dissertation addresses these challenges by proposing an innovative approach interconnecting independent IoT architectures to share resources and advanced synthetic data generation techniques to mitigate the effects of data failures.This dissertation first proposes a novel interconnected or cross-network IoT architecture, facilitating communication between independent IoT systems. This architecture enables seamless data sharing and fusion across systems, creating a robust framework for addressing data failures. The proposed architecture emphasizes flexibility, modularity, and interoperability, incorporating multiple logical layers such as perception, network, data fusion, and security layers. By interconnecting standalone/independent IoT systems, the framework reduces the reliance on individual network components and enhances overall system resilience. The architecture is designed to harmonize heterogeneous IoT networks, leveraging shared data resources to address single-point failures and reduce the need for redundant sensor deployments. To complement this architecture, the research further extends its study by developing advanced synthetic data generation techniques based on K-Nearest Neighbors (KNN) combined with Iterative Principal Component Analysis (IPCA). These methods leverage the proposed cross-network data fusion framework to create reliable synthetic data for addressing missing or unreliable datasets. Two distinct data fusion methods are presented: (1) fusion of the same feature type across networks and (2) fusion of highly correlated feature types. These approaches enable the system to compensate for missing data by utilizing information from other IoT networks sensing similar or related features. The proposed methods eliminate the limitations of traditional techniques reliant on historical or redundant data from the same network, offering a more robust and adaptable solution for dynamic IoT environments. Comprehensive experimentation validates the effectiveness of the proposed approaches using real-world IoT datasets collected from diverse geographical locations and environmental conditions. Results demonstrate that the KNN+IPCA method significantly outperforms state-of-the-art machine learning, statistical, and probabilistic approaches, achieving lower Root Mean Square Error (RMSE) values across all tested scenarios. Furthermore, the integration of the proposed cross-network architecture with synthetic data generation techniques enhances data reliability and system adaptability, even in highly heterogeneous and failure-prone environments. Thus, this dissertation advances IoT systems by addressing data failure challenges, enabling operational integrity, resource optimization, and accurate decision-making. By enhancing system resilience, this dissertation paves the way for robust, scalable, and sustainable IoT ecosystems.