Botnet Detection with Deep Neural Networks Using Feature Fusion

With the vast popularity of IoT (Internet-of-Things), cloud computing and edge computing, botnet attacks are flourishing nowadays. Meanwhile, deep learning-powered models are widely deployed to secure the network and applications. However, deep learning-based botnet detection is a challenging problem due to its extensive network traffic volume, complex feature engineering and the lack of the benchmark dataset for evaluation. With the aim of improving the performance of botnet detection, this paper firstly designs a feature extraction method by using the effective payload from each network packet. Then, a feature selection algorithm is designed based on the comparison and trade-off on the length of the extracted packets and the trained models’ performance. By choosing a reasonable number of packets and an appropriate length of bytes as feature vectors, a deep learning model is designed and evaluated for botnet detection. The experimental results prove that the designed deep neural network achieves 98% accuracy with low cost.