A Hybrid Classification Approach of Network Attacks using Supervised and Unsupervised Learning

The increasing scale and sophistication of network attacks have become a major concern for organizations around the world. As a result, there is an increasing demand for effective and accurate classification of network attacks to enhance cyber security measures. Most existing schemes assume that the available training data is labeled; that is, classification is based on supervised learning. However, this is not always the case since the available real data is expected to be unlabeled. In this paper, this issue is tackled by proposing a hybrid classification approach that combines both supervised and unsupervised learning to build a predictive classification model for classifying network attacks. First, unsupervised learning is used to label the data available in the dataset. Then, different supervised machine learning algorithms are utilized to classify data with the labels obtained from the first step and compare the results with the ground truth labels. Moreover, the issue of the unbalanced dataset is addressed using both over-sampling and undersampling techniques. Several experiments have been conducted, using the NSL-KDD dataset, to evaluate the efficiency of the proposed hybrid model and the obtained results demonstrate that the accuracy of our proposed model is comparable to supervised classification methods that assume that all data is labeled.