A Deep Learning Model for Network Intrusion Detection with Imbalanced Data

With an increase in the number and types of network attacks, traditional firewalls data encryption methods can no longer meet needs current security. As a result, intrusion detection systems have been proposed to deal with threats. The mainstream algorithms are aided machine learning but problems low rates need for extensive feature engineering. To address issue accuracy, this paper proposes model traffic anomaly named deep (DLNID), which combines attention mechanism bidirectional long short-term memory (Bi-LSTM) network, first extracting sequence features through convolutional neural (CNN) then reassigning weights each channel mechanism, finally using Bi-LSTM learn features. In public sets, there serious imbalance generally. issues, employs method adaptive synthetic sampling (ADASYN) sample expansion minority class samples, eventually form relatively symmetric dataset, uses modified stacked autoencoder dimensionality reduction objective enhancing information fusion. DLNID is end-to-end model, so it does not undergo process manual extraction. After being tested on benchmark dataset NSL-KDD, experimental results show that accuracy F1 score better than those other comparison methods, reaching 90.73% 89.65%, respectively.