Dos attack forecasting: A comparative study on wrapper feature selection

Abstract

Today, individuals, business and public administrations are internet dependent. This strong dependence creates one of the important sources of threats. Among these threats, the famous Dos attack. The costs of downtime, outages and failures caused by these attacks are very important. Protecting and preventing these threats by using the conventional tools present important limits (cannot predict in real-time when, where, and how the new forms of these Dos attacks occur). To deal with these limits, cybersecurity systems based on machine learning models can analyze patterns and learn from them to forecast and prevent Dos attack. One of the key process which ensures the efficiency of these forecasting systems is feature selection. In this context, we paid particular attention to one of the efficient feature selection methods used in forecasting cybersecurity systems: Wrapper based-feature. To find the best subset of dos attack features and to optimize the accuracy of these systems, we present a comparative study between different wrapper methods applying to the dos attack forecasting. This investigation shows that a wrapper approach based on a genetic algorithm improves the forecasting accuracy more than other wrapper processes.