Enhanced Threat Forecasting in Smart Intrusion Detection Systems with Attention CNN-BiLSTM

This study presents a new deep learning approach for identifying network security breaches; it blends CNNs with Bidirectional Long Short-Term Memory (BiLSTM) networks. To improve identification accuracy, the proposed model employs convolutional neural networks (CNNs) to extract spatial information and long short-term memory (BiLSTMs) to capture temporal connections. Using the UNSW-NB15 dataset, the model's accuracy in binary classification was 98.76%, and in multi-class classification, it was 85.50%. When compared to industry-standard models like RNN, GRU, and FNN, this one performs better. Important preprocesses, including feature selection, data normalization, and handling missing values, enhanced speed. Training became more efficient with hyperparameter modification and early termination, and sequential pattern recognition was enhanced by BiLSTM's bidirectional feature. The results show that the CNN-BiLSTM designs work well to enhance intrusion detection systems, which means that cybersecurity solutions will be more secure..