In Computational intelligence and neuroscience
In order to address the false alarm detection problem caused by the inability to identify the transgression scene pages in the process of horizontal transgression detection, this study proposes a deep learning-based LSTM-AutoEncoder unsupervised prediction model. The model uses long short-term memory network to build AutoEncoder, extracts text features of page response data of horizontal transgression scenario, and reconstructs text features to restore. Meanwhile, it counts the error between the restored result and the original page response, judges whether the detection result of horizontal transgression is false alarm according to the error threshold of unknown page, and tests the effectiveness of the model effect under real business data by comparing it with other two algorithms, one-class SVM and AutoEncoder, which provides security for enterprise network business. The results show that the LSTM-AutoEncoder model achieves a more balanced index in terms of accuracy, precision, recall, and F1-score in the case of MAE, which is 0.3% more and 0.2% more than the case of MSE in terms of recall and accuracy. It is concluded that the LSTM-AutoEncoder model is more in line with the real business requirements, and the simple model architecture selected for this study can reduce the complexity of the model, speed up the prediction time of the model in the application phase, and improve the performance of the detection software. This indicates that this study has some application prospects in network security.
Lin Shi, Yang Ma, Lu Yan, Chen Liquan