Evaluating Deep Learning Models for Network Intrusion Detection: A Comparative Analysis

The exponential rise of the Internet of Things has opened up immense opportunities. It also increases network security risks. If attacks are not monitored or prevented at the early stage, they can create acute losses to enterprises, industry, and personnel. Deep Learning and Machine Learning based intelligent intrusion detection system are the answers to this challenge. This paper proposes an Intrusion Detection System (IDS) using popular DL and ML algorithms. We used the KDDcup99 dataset to compare supervised ML algorithms to DL algorithms. We evaluated the algorithm’s performance by comparing metrics such as accuracy, precision, F1-score, as well as sensitivity (TPR), and the rate of false positives (FPR). Our experiment showed that DL-based algorithms are more accurately predicted than ML algorithms regarding all performance parameters used in this paper. BiLSTM performed highest in all measures compared to other algorithms and accuracy at 98%. These results illustrate DL algorithms’ strong capability and potential over ML to detect network attacks.