A blockchain and stacked machine learning approach for malicious nodes’ detection in internet of things

The Internet of Things (IoT) network is extremely useful in many different fields, such as smart cities, the military, healthcare, business, and agriculture, among others. The security of IoT networks has long been a major concern. The nodes of the IoT network are openly accessible and are mostly deployed in hostile environments. Therefore, they are easily exposed to different attacks. It is important to identify malicious nodes and attacks present in the network. The majority of traditional IoT security mechanisms are either centralized or rely on third parties. Thus, they are vulnerable to a single point of failure. Moreover, Machine Learning (ML) based attack detection approaches have low detection accuracy and performance as compared to the deep learning approaches. Besides, the single learner algorithms have intrinsic limitations that directly impact the attack detection systems’ performance. To address the aforementioned issues, a Blockchain Ensemble stacked Machine Learning (BEML) approach has been proposed in this article. The BEML approach is made up of three modules: blockchain, InterPlanetary File System (IPFS) and attack detector. The blockchain module registers network nodes, authenticates data analysts, revokes network nodes, and stores data hashes and credentials’ information of nodes in a secure way. The IPFS module stores the data and generates a unique hash against it. Later on, the hash is used to access and download data from the IPFS. In the third module, the raw data is processed, normalized and balanced using the MinMax scalar and Synthetic Minority Oversampling Technique (SMOTE). Moreover, the single learner ML algorithms: linear discriminant analysis, decision tree, perceptron and ridge, are combined in such a way that they compensate for the weaknesses of each other and result in a stacked ML model. The resulting model has better performance and detection accuracy as compared to the single learner algorithms. This stacked ML model is used to detect and classify the Denial of Service (DoS) attacks present in the network. Based on the predicted attacks, the malicious nodes are identified and their registration is revoked from the network. Finally, for evaluating the the proposed BEML approach’s efficiency, simulations are performed. The results, theoretical analysis and formal security analysis indicate that the BEML approach effectively stores data, efficiently detects attacks and ensures the security of the IoT network.