This paper explores the application of deep learning in intrusion detection systems (IDSs) for time series data. The authors propose a new architecture that combines convolutional neural networks (CNNs), long short-term memory networks (LSTMs), and attention mechanisms to forecast malicious packets in real-time. The dataset used is the UNSW-NB15, which contains 2.5 million network packets and 49 corresponding labels. The proposed model is evaluated using various metrics such as F1 score and AUC, and it achieves an F1 score of 83% for predicting the existence of an attack in the next packet, which is comparable to real-time IDS classification. The study also includes an ablation study to validate the architectural choices and concludes with a discussion on the trade-offs between computational complexity and accuracy. The main contributions of the paper include extending existing IDS systems with proactive prediction capabilities, comparing the proposed approach with state-of-the-art methods, and conducting thorough experiments to optimize hyperparameters. The results demonstrate the effectiveness of the proposed model in proactive intrusion detection, highlighting its potential for enhancing cybersecurity measures.This paper explores the application of deep learning in intrusion detection systems (IDSs) for time series data. The authors propose a new architecture that combines convolutional neural networks (CNNs), long short-term memory networks (LSTMs), and attention mechanisms to forecast malicious packets in real-time. The dataset used is the UNSW-NB15, which contains 2.5 million network packets and 49 corresponding labels. The proposed model is evaluated using various metrics such as F1 score and AUC, and it achieves an F1 score of 83% for predicting the existence of an attack in the next packet, which is comparable to real-time IDS classification. The study also includes an ablation study to validate the architectural choices and concludes with a discussion on the trade-offs between computational complexity and accuracy. The main contributions of the paper include extending existing IDS systems with proactive prediction capabilities, comparing the proposed approach with state-of-the-art methods, and conducting thorough experiments to optimize hyperparameters. The results demonstrate the effectiveness of the proposed model in proactive intrusion detection, highlighting its potential for enhancing cybersecurity measures.