المستخلص: |
The importance of computers’ security appears as a high priority mission in the world of the Internet with the wide spreading of the size of computers network and an ever-growing number of applications available on it against BotNet attacks in different fields. Moreover, Intrusion Detection Systems (IDS) and Intrusion Prevention Systems (IPS) play a vital role in prevention and defence in a confrontation of the propagation ways that Botnets attacks spread in. Since the Internet is so prevalent in human life, cyber threats, particularly attack detection, are a challenge for cyber security research. Intrusion detection systems (IDSs) are key network topology organizations aimed at protecting the integrity and availability of critical assets in protected systems. Although many supervised and unsupervised learning approaches from a machine learning field and model recognition have been used to increase the effectiveness of IDSs, many redundant and irrelevant features in high-dimension network anomaly detection datasets are still problematic. In this research, we suggest a novel approach integrating the benefits of Minimum Redundancy Maximum Relevance Feature Selection with an ensemble classifier based on neural networks and Random Forest (RF), which can be capable to classify together familiar and uncommon kinds of attacks with great accuracy and efficiency. Our ensemble approach obtained a 98.09 percent detection rate and 99.06 percent accuracy values. We compared other approaches to our approach on the same dataset and found that the strategy proposed exceeded existing techniques. The suggested method was checked and analyzed using real data sets of network traffic to validate the solution's performance.
|