Network Anomaly Detection With Temporal Convolutional Network and U-Net Model

Anomaly detection in network traffic is one of the key techniques to ensure security future networks. Today, importance this topic even higher, since growing and there a need have smart algorithms, which can automatically adapt new conditions, detect threats recognize type possible attack. Nowadays, are lot different approaches, some them reached relatively sufficient accuracy. However, majority works being tested on old datasets, do not reflect current conditions it leads overfitted results. This caused by high redundancy data because they fail performance latest methods real-world anomaly applications. In work, we applied couple based convolutional neural networks: U-Net Temporal for attack classification. We trained evaluated dataset KDD99 modern large-scale CSE-CIC-IDS2018. According results, with LSTM has achieved accuracy 92% 97% CSE-CIC-IDS2018 respectively, model 93% 94% respectively. Additionally, utilized focal loss function Long Short-Term Memory model, positive effect class imbalance time-series data. showed, that combination give higher compared other architectures work also proved, easily overfit during training achieve good results testing set, but at same time, these so successful more complex actual