Analysis of Feature Selection Approaches in Large Scale Cyber Intelligence Data with Deep Learning

Abstract

The size of the network systems that grows day by day causes the attack density and types to increase. Detection of these attacks within the network is one of the main problems of network security. Intrusion detection systems are an approach developed to deal with this problem. Large data processed in intrusion detection systems also brings complexity. This study includes examining 6 different attribute selection algorithms and comparing the performance of these algorithms in classification models to eliminate the complexity in data sets. These performances were analyzed with Deep Learning models applied on the open access CICIDS2017 data set. During this process, the test results of the algorithms were compared both among themselves and with the original form of the data set. During implementation, the number of attributes in the dataset was reduced from 78 to 25 for multiple classification and to 8 for binary classification. The success rates obtained are over 92% in all applications.