A Performance Evaluation of Neural Networks for Botnet Detection in the Internet of Things

Author(s)

Guimaraes, Lucas C. B. ; Couto, Rodrigo S.

• Document type: Journal article
• Source: Journal of Network and Systems Management; v. 32, n. 4, p. 24-pg., 2024-10-01.

Abstract

IoT (Internet of Things) devices are fundamental to sectors such as smart homes and cities, industry 4.0, and smart grids. Despite the benefits brought by IoT, the existence of billions of devices with limited computing resources makes them ideal targets for botnets. Thus, multiple proposals have been made to detect this type of attack. However, comparing different proposals is difficult since they apply varied preprocessing methods, use different algorithms and hyperparameters, and consider distinct evaluation metrics. This paper implements and compares the performance of eight neural network architectures applied to the BoT-IoT and N-BaIoT datasets, which contain botnet attacks and labeled IoT network traffic. The models' accuracy, precision, and recall are measured, as well as the loss during model training. Afterward, the models' throughput on an edge environment is evaluated using a typical edge device, an NVIDIA Jetson Nano, while also implementing quantization and evaluating its impact on model accuracy. The results show that, after hyperparameter tuning, several BoT-IoT models exceed 99% accuracy while most N-BaIoT models surpass 80% accuracy. However, the throughput results show that the best-performing model might not scale in environments composed of a large number of IoT devices, even considering the influence of 8-bit quantization. (AU)

• FAPESP's process: 23/00811-0 - EcoSustain: computer and data science for the environment
• Grantee: Antonio Jorge Gomes Abelém
• Support Opportunities: Research Projects - Thematic Grants
• FAPESP's process: 23/00673-7 - Distributed intelligence in communications networks and in the internet of things
• Grantee: Nelson Luis Saldanha da Fonseca
• Support Opportunities: Research Projects - Thematic Grants