A Deep Learning Framework for Detecting Cyber Vulnerabilities in IOT-Based Data Exchange Systems

Abstract

The increased use of the Internet of Things (IoT) has enabled improved data exchanges that span various fields, such as healthcare, industrial systems, smart cities, and transportation. However, the growth of the IoT paradigm has presented greater challenges in securing data exchanges due to the increased attack surfaces created by connected devices. Threats such as Distributed Denial-of-Service (DDoS), malware, botnets, brute-force, and web intrusions can target IoT-based systems, thus compromising the confidentiality, integrity, and availability of data. Current cybersecurity solutions are inadequate in detecting advanced persistent threats due to their limited ability to understand the temporal relationships between network traffic features. This work proposed a Temporal Fusion Transformer (TFT)-based approach to identifying cyber vulnerabilities in IoT-based data exchange systems. The publicly available Edge-IIoTset dataset was used for benchmarking purposes. The data was preprocessed, including cleaning, removing duplicates, handling missing values, encoding, Min-Max normalization, feature selection, balancing class distributions using Synthetic Minority Oversampling Technique (SMOTE), and sequential data preparation. The proposed TFT was trained and validated using the Adam optimizer and categorical cross-entropy loss, with a 70:15:15 training/validation/testing split and early stopping. The model achieved a testing accuracy of 93.4%, precision of 93.8%, recall of 92.6%, an F1-score of 92.8%, 96.8% ROC-AUC, and a Matthew’s correlation coefficient (MCC) of 0.912. Additionally, the framework registered a false positive of 4.2%, 34.5ms detection latency, and 0.52ms inference time per batch. TFT surpassed traditional and state-of-the-art deep learning counterparts in detecting different categories of IoT-based cyberattacks. The work concluded that the proposed framework could offer a reliable and scalable solution to securing IoT-based data exchange systems.

DOI: https://doi.org/10.5281/zenodo.21724806