Multiclass Classification of Attacks in APCS by Applying SW Graph

Based on the UNSW-NB 15 training dataset, a preliminary exploratory data analysis was conducted, including working with categorical and numerical attributes, and correlating them with each other to reduce data dimensionality and filter out noise. Two approaches to the problem of classifying multi-class attacks were considered: "One against All" and "All against All". Based on these approaches, the constructed machine learning models are evaluated using precision, recall, f1measure metrics. As a result of the study, an approach for anomaly detection and attack classification in the case of a small training sample using the construction of SW-graph is proposed. The algorithm, workflow and main advantages of using graph method are described.