Vol. 8, Issue 2 (2019)

Author(s):
Priyanka

Abstract:
Cyberattacks that have the potential to bring down a network are continuously being sought after in the field of network security. Malicious acts on the network are also expanding quickly due to the unexpected creation and increased use of the Internet. An effective intrusion detection system (IDS) is necessary to stop unwanted access to network resources in order to identify anomalies in the network and safeguard data. Recently, a number of notable approaches have been put forth as a cure-all for intrusion detection, but it is still difficult to construct a secure system because attackers frequently alter their tactics to get around the system's security measures. In this research, the categorization of data into normal or intrusive categories was accomplished through the application of machine learning (ML) classifiers. A diverse set of classifiers was utilized in this study, encompassing logistic regressions (LR), extra-tree classifiers (ETC), Decision trees (DT), logistic support vector machines (SVM), random forests (RF), Naive Bayes (NB), multi-layer perceptron’s (MLP), and K-nearest neighbors (KNN). Four feature subsets from the NSL-KDD dataset were used in the study to evaluate the model's efficacy. A thorough pre-processing of the data was carried out, which included deleting unnecessary attributes from the dataset. This was a critical step because it acknowledged that an intrusion detection system’s dimensional aspect are closely related to its effectiveness. The empirical findings revealed that KNN exhibited a performance exceeding 99 percent across all attack classes when applied to various feature subsets. Consequently, through the strategic removal of unnecessary features, the proposed model not only mitigates computational complexity but also attains a notable high prediction accuracy rate.