An Intelligent Botnet Detection System for IoT Using Neural Networks and an Enhanced Moth Search Optimize

Abstract

Botnets, distributed networks of compromised devices under remote control, continue to pose a serious cybersecurity threat, as they are challenging to detect with traditional methods because of their evasion capabilities. The evolving nature of botnets demands stronger detection systems that can effectively detect malicious traffic patterns. To confront this problem, we introduce a new botnet detection framework called BDS (Botnet Detection System), aimed at improving the accuracy of detection and reducing the number of false positive results. We integrate the Enhanced Moth Search Algorithm (EMSA) with Multi-Layer Perceptron (MLP) for the enhancement of training of the neural network model, titled EMSA-MLP. For this, we test our model with a representative Bot-IoT dataset with diverse botnet attack scenarios in terms of separating honest and malicious traffic with the help of efficient optimization by EMSA. To assess the proposed EMSA-MLP, the Bot-IoT benchmark dataset was used, containing a variety of botnet attack methods. The detection performance of the model was demonstrated to be excellent. The model achieved a high detection performance in identifying botnet attacks, with an accuracy rate of 97.09%. They also preserved a good accuracy of 91.59%, and their false alarm rate was kept low at 0.0291. Overall, compared to some common classifiers: random forest, decision tree, and base MLP-this model did pretty well. It also outperformed relatively newer and more complex models. This architecture has achieved good accuracy, while it has not made the growth of IoT settings very large, and it has been formulated to be used by devices. This model provides high accuracy without burdening the IoT infrastructure, so it is applicable to devices with restricted capabilities. It is suited for practical applications, as it can adapt to different types of data.