Sin-Cos-bIAVOA A new feature selection method based on improved African vulture optimization algorithm and a novel transfer function to DDoS attack detection

Internet of Things (IoT) services and devices have raised numerous challenges such as connectivity, computation, and security. Therefore, networks should provide and maintain quality services. Nowadays, Distributed Denialof-Service (DDoS) attack is the most important network attacks according to recent studies. Among the variety of DDoS detection methods, Machine Learning (ML) algorithms have attracted researchers. In ML, the selection of optimal subset of features can have a significant role to enhance the classification rate. This problem called the feature selection problem is in the class of NP-hard problems and exact algorithms cannot obtain the best results in acceptable time. Therefore, approximate algorithms like meta-heuristic algorithms are employed to solve the problem. Since these algorithms do not search all solution space, they fall in local optima and provide a premature convergence rate. Several methods have been introduced so far to address these challenges but researchers try to find new strategies for enhancing the performance of methods. In this study, a binary Improved African Vulture Optimization Algorithm (Sin-Cos-bIAVOA) is proposed to select effective features of DDoS attacks. The method applies a novel compound transfer function (Sin-Cos) to increase exploration. To select the optimal subset of features, Gravitational Fixed Radius Nearest Neighbor (GFRNN) is employed as the classifier in the method. Moreover, AVOA is improved in three phases including exploration, balancing exploration and exploitation, and exploitation phases. Hence, Sin-Cos-bIAVOA explores promising areas to achieve the best solution and avoid the local optima traps. The proposed method's performance is compared with some recent stateof-the-art in two datasets, CIC-DDOS2019 and NSL-KDD for the DDoS attack detection. The experiment results show that the proposed method achieves the minimum feature selection rate (0.0184) with the high average accuracy (99.9979%), precision (99.9979%), recall (100.00%), and F-measure (99.9989%) compared with competitors in the first scenario with 1% attack rate in CIC-DDOS2019 dataset. In addition, the results of Friedman test based on fitness functions indicate that Sin-Cos-bIAVOA has the first rank among comparative algorithms. The source code of Sin-Cos-bIAVOA is publicly available at https://www.mathworks.com/matlabcentral/fileexchange/129409-sin-cos-biavoa-a-new-feature-selection-method.