AI and Blockchain-Assisted Secure Data-Exchange Framework for Smart Home Systems

Abstract

The rapid expansion of the Internet of Things (IoT) on a global scale has facilitated the convergence of revolutionary technologies such as artificial intelligence (AI), blockchain, and cloud computing. The integration of these technologies has paved the way for the development of intricate infrastructures, such as smart homes, smart cities, and smart industries, that are capable of delivering advanced solutions and enhancing human living standards. Nevertheless, IoT devices, while providing effective connectivity and convenience, often rely on traditional network interfaces that can be vulnerable to exploitation by adversaries. If not properly secured and updated, these legacy communication protocols and interfaces can expose potential vulnerabilities that attackers may exploit to gain unauthorized access, disrupt operations, or compromise sensitive data. To overcome the security challenges associated with smart home systems, we have devised a robust framework that leverages the capabilities of both AI and blockchain technology. The proposed framework employs a standard dataset for smart home systems, from which we first eliminated the anomalies using an isolation forest (IF) algorithm using random partitioning, path length, anomaly score calculation, and thresholding stages. Next, the dataset is utilized for training classification algorithms, such as K-nearest neighbors (KNN), support vector machine (SVM), linear discriminate analysis (LDA), and quadratic discriminant analysis (QDA) to classify the attack and non-attack data of the smart home system. Further, an interplanetary file system (IPFS) is utilized to store classified data (non-attack data) from classification algorithms to confront data-manipulation attacks. The IPFS acts as an onsite storage system, securely storing non-attack data, and its computed hash is forwarded to the blockchain’s immutable ledger. We evaluated the proposed framework with different performance parameters. These include training accuracy (99.53%) by the KNN classification algorithm and 99.27% by IF for anomaly detection. Further, we used the validation curve, lift curve, execution cost of blockchain transactions, and scalability (86.23%) to showcase the effectiveness of the proposed framework.