Assessing the Efficacy of SVM Kernel Types for Detecting Generic Attacks in Cloud Environments: A Meta-Heuristic Perspective

Any form of abnormality or network attack that doesn’t fall into one of the conventional categories, such as Reconnaissance, Shellcode, Worms, DoS (Denial of Service), or probing, is referred to as a Generic attack. Rather, it stands for attacks that are difficult to classify into preexisting attack classes or may not have a clear classification. As such, compared to clearly defined attack types, effectively recognizing and classifying them may be more difficult. The literature by the scientific community covered a range of generic attack methods. Real-time resolution of generic attacks poses significant risks. A few crucial features of each of these attack kinds are recognized in order to enhance cloud network security instantly. In this research, the essential features from the UNSW-NB15 dataset is extracted, utilizing the particle swarm optimization (PSO) method. Using these features as the input, the four support vector machine kernel functions—radial basis function, linear, polynomial, and sigmoid—help the suggested technique be easily expanded for the detection of novel and undetected generic attacks. A balanced generic network dataset was selected and employed in the model-building process. For the purpose of developing the model and assessing its performance, a 10-fold cross-validation approach was used, from which all possible performance values were obtained and shown. The results of the experimental study unequivocally demonstrate that the Radial Basis support vector machine kernel function provides the best performance in terms of identifying both apparent and undetected generic attacks. A comparison analysis and assessment of the SVM kernels’ performances were conducted for the purpose of detecting generic network intrusion attacks.