An Abnormal Traffic Detection Model Combined BiIndRNN With Global Attention

As time series data with internal correlation, networks traffic can be used for abnormal detection using Recurrent Neural Network (RNN) and its variants, but existing models are difficult to calculate in parallel, gradient explosion or vanishing easily occurs. To address this problem, we propose a Bidirectional Independent (BiIndRNN) parallel computation adjustable gradient, which extract the bidirectional structural features of by forward backward input capture spatial influence flow. establish dependencies on moments traffic, model combining Global Attention (GA) BiIndRNN is proposed pay more attention containing essential information. Taking UNSW-NB15 dataset as object, GA expression packets feature vector derived, fusion, well loss calculation, performed multiple fully connected layers. The experimental results show that, compared traditional deep shallow machine learning other state-of-the-art technologies, our GA-BiIndRNN converges faster, accuracy, precision, F1 scores all above 99%, false positive rate (FPR) close 0.36%, effectively identify normal malicious network activities. These provide theoretical basis rapid implementation protective measures.