Attentional payload anomaly detector for web applications

Abstract

Nowadays web applications influence people deeply and become popular targets of attackers. The payload anomaly detection is an effective method to keep the security of web applications but requires proper features which takes a lot of time and effort for experts and researchers to design. Utilizing the deep learning techniques for the detection is a solution to the feature design problem because deep learning models can learn features during the training process and achieve great performances. However, current deep learning payload detection models have their limit on processing long sequences, which reduces the detection performance. And due to the intricate data processing, the results produced by the models are unconvincing. In this paper, we proposed an attentional recurrent neural network (RNN) model for the payload detection, called ATPAD. With the attention mechanism, ATPAD generates effective features for the detection tasks and provides a visualized way to verify detection results. The experiment results show that our proposed model not only achieves high detection rates and low false alarm rates, but also produces understandable results.