Improved Gated Recurrent Unit Based Network Traffic Anomaly Detection

The detection of unpredicted events or rare patterns in traffic data plays an important role in intelligence traffic management, that offers alerts of incidents. To address the problem of less accuracy in industrial network traffic anomaly detection, in this manuscript developed the Improved Gated Recurrent Unit (GRU) for multi-type anomaly detection. The pre-processing included label encoding and non-string numerical process. The Synthetic Minority Over-Sampling Technique with Edited Nearest Neighbors (SMOTE-ENN) technique is utilized for balancing the data by oversampling and under sampling. The improved GRU is assigned for extracting linear, non-linear and temporal features from network traffic data. These features are given to classifier developed based on Softmax for obtaining high performance detection on traffic attacks. The efficiency is evaluated based on UNSW-B15 dataset with the detection accuracy of 0.97 for DoS attacks, 0.96 for Fuzzers attacks and 0.94 for Worms attacks while compared to existing algorithms like conventional GRU for network traffic anomaly detection.