An improved binary manta ray foraging optimization algorithm based feature selection and random forest classifier for network intrusion detection

Abstract

• Proposed a binary version of manta ray foraging optimization algorithm for binary optimization problems. • Applied the proposed binary manta ray foraging optimization algorithm for feature selection using NSL-KDD and CICIDS2017 network traffic datasets. • Developed network intrusion detection model with the selected features based on random forest classifier. • Examine performance analysis and comparison of the presented model with GA, PSO, GWO, and GOA using Accuracy, Recall, Precision, F-measure, and execution time. • Statistical justification of the results achieved compared to that of GA, PSO, GWO, and GOA using t -test. The growth within the Internet and communications areas have led to a massive surge in the dimension of network and data. Consequently, several new threats are being created and have posed difficulties for security networks to correctly discover intrusions. Intrusion Detection System (IDs) is one amongst the foremost essential events for security arrangements in network environments, and it is commonly applied to spot, track, and detect malevolent threats. Detecting intruders using metaheuristics and machine learning methodologies in recent trend offers improved discovery rate. Therefore, this paper presented an intrusion detection model using an improved Binary Manta Ray Foraging (BMRF) Optimization Algorithm based on adaptive S-shape function and Random Forest (RF) classifier. The BMFR is envisioned to identify the most relevant features and remove redundant and irrelevant ones from the intrusion detection datasets. Furthermore, the RF is used for feature evaluation and to build the intrusion detection model. The proposed method was validated and compared with other methods using two IDs benchmark datasets, which include NSL-KDD and CIC-IDS2017 datasets. The result indicates that the presented model selected 38 features with 99.6% precision, 94.3% recall, 96.9% f-measure, and 99.3% accuracy for the CIC-IDS2017 dataset. Moreover, for the NSL-KDD dataset, the presented model selected 22 features with 96.8%, 96.2%, 96.5%, and 98.8% for precision, recall, F-measure, and accuracy. In addition, a statistical significance test reveals a significance difference between the presented model and the compared methods in terms of F-measure.