基于改进双向长短期记忆神经网络的恶意请求检测
机器学习
2020-11-13 v4 网络与互联网体系结构
摘要
检测和拦截恶意请求是网络安全中对抗攻击最广泛使用的方式之一。大多数现有的检测方法,包括黑名单字符匹配和机器学习算法,均已被证明易受复杂攻击的影响。为解决上述问题,需要一种更通用且严谨的检测方法。本文将恶意请求检测问题建模为时间序列分类问题,并提出一种新颖的深度学习模型,即卷积神经网络-双向长短期记忆-卷积神经网络(CNN-BiLSTM-CNN)。通过连接卷积层的浅层与深层特征图,在更细粒度的功能上提升了恶意特征提取能力。在 HTTP 数据集 CSIC 2010 上的实验结果证明了所提方法相较于现有最优方法的的有效性。
引用
@article{arxiv.2010.13285,
title = {Malicious Requests Detection with Improved Bidirectional Long Short-term Memory Neural Networks},
author = {Wenhao Li and Bincheng Zhang and Jiajie Zhang},
journal= {arXiv preprint arXiv:2010.13285},
year = {2020}
}
备注
Mistakes occupy in the experimental results thus was rejected by ICICS. For preciseness we have to withdraw the paper. The experimental results are self-contradictory. The results of RNN-IDS in Table 1. FPR = 100% = FP/(FP+TN), so TN should be 0. recall = 100% = TP/(TP+FN), so FN should be 0. Therefore, precision=TP/(TP+FP)=50%, given that FP=TP. But table 1 shows that precision is 0.6967