Feature level fusion of multi-source data for network intrusion detection

Abstract

The generation of data, collecting, and refining in computer networks have increased exponentially in recent years. Network attacks have also grown in prevalence with this proliferation of data and are now an inherent issue in complicated networks. Current network intrusion detection systems (NIDS) have significant issues with regard to anomaly detection. Several machine learning classification approaches are used to create NIDSs, but they are not sufficiently sophisticated to reliably detect complicated or synthetic attacks, especially if working with a lot of multi-scale data. Data fusion has been used in network intrusion detection to address these issues. For network intrusion detection, we suggested a multi-source data fusion technique in this research, which combines specific features from two datasets to produce a single dataset. Also, a machine learning classifier with fewer parameters is utilized for the fused dataset. The random forest shows the best classification accuracy compared to others in this work. For the normal classification, model accuracy is 92.8%, and the proposed fusion model showed 97.3% accuracies. Furthermore, the findings show that, when compared to other cutting-edge techniques, the suggested model is substantially more effective in detecting intrusions.