Research on Network Traffic Anomaly Detection Method Based on Autoencoders

With the continuous expansion of the network scale, network attacks are becoming increasingly frequent, drawing more attention to network security issues. Network traffic anomaly detection plays a crucial role in preventing network attacks. However, traditional detection methods often face challenges with large-scale network traffic data and low detection accuracy. To address this, a network traffic anomaly detection method based on autoencoders is proposed. This method reconstructs network traffic using an autoencoder neural network, and determines anomalies based on the size of the reconstruction error. To overcome the challenge of selecting an appropriate reconstruction error threshold that balances recall and precision, an adaptive reconstruction error threshold is proposed. It is selected based on the distribution of reconstruction errors in the training set data, given the known a priori probability of anomalies and the minimum precision requirement, with the aim to improve the recall rate and mitigate the underreporting of anomalous network traffic. Experimental results demonstrate that the proposed method can effectively detect abnormal network traffic with a high recall rate. This method can serve as a valuable reference for further research in network traffic anomaly detection.