SMOTE: Class Imbalance Problem In Intrusion Detection System

In the last decade, the size of data is exponentially increasing, making the data classification process more challenging, especially when the data has an imbalanced distribution of classes. Which means that one majority class is having more instances than other classes. In this case, standard classifiers tend to classify all examples as a majority class and can completely ignore the minority class. For this problem, researchers proposed multiple solutions, and most of these efforts are focusing on binary-class problems. However, in the case of multi-class dataset, it is more difficult to deal with the data. Hence, multi-class classification in imbalanced dataset remains a significant topic of research. In this work, by examining a recent multi-class dataset called CICIDS2017, which is related to Intrusion Detection System (IDS) domain, and has been analyzed for similar issues. We examine this dataset to test one of the popular approaches to solve the imbalance problem in multi-class datasets, which is the Synthetic Minority Over-sampling Technique (SOMTE). Based on the result, we conclude that SMOTE has increased the performance of classifying classes of the examined imbalanced dataset.