A Review of Benchmark Datasets and its Impact on Network Intrusion Detection Techniques
C Haripriya, Prabhudev Jagadeesh · 2022
With the advancements in Internet technology and emergence of new devices and network architectures, Cyber-attacks are becoming more sophisticated. As always with new technology, comes new threats. The main challenge is to accurately detect these intrusions and alert the network administrator to prevent these attacks. Deep learning techniques have proven to be more effective when compared to shallow machine learning methods. Intrusion detection can be classified into several types based on architecture, implementation, algorithm type etc. Dataset plays a vital role in the performance analysis of the model. Quality of the dataset adversely affects the training and would to lead to incorrect and inconsistent model results. The availability of datasets for conducting experiments for IDSs (Intrusion Detection Systems) continues to be a problem. Many publicly available datasets do not always accurately reflect data from the real world. On the other hand, because they are made available to the public, they make it possible for researchers to do similar benchmarking. On the basis of self-generated datasets, some experiments are being conducted. They may be better suited to a particular study group's needs, although privacy issues may arise. This paper discusses in detail, the different datasets used in Network Intrusion Detection. Different metrics that can be used in evaluating intrusions are also discussed. Impact of the datasets in model accuracy, challenges and future directions in carrying out research in Network intrusion detection are also addressed.