Efficient Software Vulnerability Detection with Minimal Data Size in 5G-Iot

Abstract

This research addresses the security challenges arising from the widespread implementation of Internet of Things (IoT) networks in the industry, a situation exacerbated by the integration of artificial intelligence (AI) and 5G mobile communication technologies. While these IoT networks enhance industrial efficiency and convenience, their vulnerability to security breaches is heightened due to nonstandard protocol stacks and open-source software that lacks thorough vetting. Traditional AI-based vulnerability detection methods, with their high memory and computational demands, are unsuitable for IoT applications. In response, this study introduces a novel approach utilizing the Sampling Threshold Optimization Strategy. This method maximizes the training data size while preserving learning accuracy, enabling swift identification of software vulnerabilities in IoT systems. Experiments conducted on a software vulnerability classification dataset reveal that the effectiveness of the models is not significantly impacted by variations in ideal data sizes. Notably, the research establishes the minimum amount of data required to train a model without compromising performance. For instance, employing the random forest model with the sampling threshold optimization technique resulted in a 97.9% improvement in latency and conserved 85.2% of memory, 1% of the data while still achieving learning accuracy that is on par with 100% of the data. This innovative method provides a practical means to enhance the security of IoT networks without the need for extensive processing power. © 2023 IEEE.