Network Intrusion Prediction Model based on Bio-inspired Hyperparameter Search

In recent years, predicting network intrusions has been a serious research concern in the academia and industry due to the expanding attack surfaces. This unending trend of threat escalation implies that more robust approaches are required to accurately predict attacks. Extant prediction models are generally affected by the choice of hyperparameters based on expert knowledge. Consequently, in this paper we present a novel approach to predict cyberattacks using bio-inspired hyperparameter search technique to generate an optimal network structure using core components of a deep neural network as chromosomes. This optimal or bio-inspired network structure is further used to derive a novel prediction model called NetBiiDenns. Furthermore, we evaluated our model on two well-known benchmark datasets, which include the CICIDS2017 and NSL-KDD datasets and achieved a prediction accuracy of 99%. From findings, our model predicts cyberattacks with high accuracy, low error and false positive rates, and significantly outperforms state-of-the-art comparative models.