Multi-stage deep investigation pipeline on detecting malign network traffic

作     者：Nasreen Fathima, A.H. Syed Ibrahim, S.P. 

作者机构：School of Computer Science and Engineering Vellore Institute of Technology Chennai India 

出 版 物：《Materials Today: Proceedings》 (Mater. Today Proc.)

年 卷 期：2022年第62卷

页      面：4726-4731页

核心收录：

学科分类：0710[理学-生物学] 0810[工学-信息与通信工程] 0711[理学-系统科学] 1205[管理学-图书情报与档案管理] 08[工学] 0805[工学-材料科学与工程（可授工学、理学学位）] 080502[工学-材料学] 0835[工学-软件工程] 0836[工学-生物工程] 0701[理学-数学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

主　　题：Decision trees 

摘      要：To detect intrusions effectively, we need vulnerable information. To collect the network traffic transactions and assess the anomaly which includes essential details related to network traffic data, we explore the datasets that have intrusion related network traffic. Based on this assessment, to predict attacks in the network transaction efficiently, we use the most upgraded benchmark datasets NSL-KDD and CIC-IDS-2017. In this work, a novel multistage deep learning model is developed to evaluate the optimal features for training. Our proposed model identifies weird patterns of network traffic with probabilistic unsupervised learning technique variational autoencoder. By pre-training the model in the first and second stage to classify the attack with fusion of random forest at the final stage, we improve the outcome of a very biased class dataset. Experimental results unveil the improvisation of the overall computational and time complexity. The proposed model accuracy measures as 99% with improved precision for CIC-IDS-2017 dataset. Class imbalance handling techniques have to be combined as future work in order to reduce the high false positive rates and accordingly increase the low false negative rates for the minority classes. © 2022