Abstract
Abstract. The rapid development of the Internet of Things (IoT) and the widespread deployment of network-connected devices have led to a significant increase in network traffic and a growing number of cyberattacks. Traditional intrusion detection methods often require the analysis of complete network flows and considerable computational resources, which limits their applicability in IoT environments characterized by constrained processing power and memory capacity. As a result, the development of efficient approaches for the early detection of cyberattacks has become an important research direction, allowing security systems to identify potential threats at the initial stage of network communication and respond to malicious activity more quickly. This study presents an approach for early cyberattack detection based on a hybrid deep learning architecture that integrates Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks. The CNN component is responsible for extracting spatial features from network traffic data, while the LSTM layers capture temporal dependencies within sequences of packets. To enhance the informativeness of traffic analysis, an extended statistical–dynamic feature set is introduced, including packet size distribution entropy, inter-packet time intervals, and the coefficient of variation. Experimental evaluation demonstrates that the proposed approach achieves higher classification accuracy compared with traditional machine learning algorithms such as Support Vector Machine (SVM), Random Forest, and Multilayer Perceptron (MLP). The analysis of the first ten packets of a network connection significantly reduces the time required to detect attacks, which is particularly important for real-time security systems. Furthermore, an optimized version of the model designed for IoT environments reduces computational load while maintaining high detection accuracy and practical applicability in modern network infrastructures.