Abstract
In the current cyber world, one of the most severe cyber threats are distributed denial of service (DDoS) attacks, which make websites and other online resources unavailable to legitimate clients. It is different from other cyber threats that breach security parameters; however, DDoS is a short-term attack that brings down the server temporarily. Appropriate selection of features plays a crucial role for effective detection of DDoS attacks. Too many irrelevant features not only produce unrelated class categories but also increase computation overhead. In this article, we propose an ensemble feature selection algorithm to determine which attribute in the given training datasets is efficient in categorizing the classes. The result of the ensemble algorithm when compared to a threshold value will enable us to decide the features. The selected features are deployed as training inputs for various classifiers to select a classifier that yields maximum accuracy. We use a multilayer perceptron classifier as the final classifier, as it provides better accuracy when compared to other conventional classification models. The proposed method classifies the new datasets into either attack or normal classes with an efficiency of 98.3% and also reduces the overall computation time. We use the CAIDA 2007 dataset to evaluate the performance of the proposed method using MATLAB and Weka 3.6 simulators.
Classification of Diabetes using Random Forest with Feature Selection Algorithm
