ALSNAP: Attention-based long and short-period network security situation prediction

The Internet of Things (IoT) and other emerging communication and network technologies enable improved efficiency, convenience, and connectivity in various domains. However, these advancements also pose significant security risks due to the increasing complexity of attacks and the lack of effective security strategies. Traditional security schemes, such as signature-based intrusion detection systems, have limited predictive power and require a considerable number of cybersecurity staff. Therefore, proactive defense techniques based on network security situation prediction are essential. Network security situation prediction aims to forecast the future state of network security based on historical data and current observations. However, current prediction methods can only capture either long-term or short-term data features, leading to unsatisfactory results. In contrast, this paper proposes a novel attention-based long and short-period network security situation prediction (ALSNAP) scheme that combines improved bidirectional long short-term memory, vector autoregression, and multi-scale convolution with branching attention. It applies an attention mechanism to optimize the prediction model and enhance its accuracy. We evaluated ALSNAP on four public network datasets, including three typical IoT datasets, and compared it with other state-of-the-art methods. Our results demonstrate that ALSNAP achieves higher prediction accuracy than other methods. We also analyzed the potential applications of our scheme to real networks and its integration into intelligent network security situation awareness systems to deliver timely and effective countermeasures.