HYPER META-HEURISTIC IMPROVED PARTICLE SWARM OPTIMIZATION BASED SUPPORT VECTOR MACHINE (HMHIPSO-SVM) FOR BIG DATA CYBER SECURITY

Cyber security in the domain of big data is regarded to be a crucial challenge for the research community. Machine Learning (ML) algorithms are recommended to be the representatives for dealing with big data security issues. Generally, Support Vector Machine (SVM) has been found successful for different problems of classification. The user has to specify the right SVM configuration much earlier, which again is cumbersome, needing the help of skilled experts in choosing the kernel function and a excessive amount of manual labor for experiments. In order to resolve this problem, a HMHIPSO-SVM classifier is proposed to be designed, whose performance is necessary for dealing with the right choice of the low-level heuristic. Hyper Meta-Heuristic Improved Particle Swarm Optimization based Support Vector Machine (HMHIPSO-SVM) framework comprises of a high-level mechanism and low-level heuristics. The high-level mechanism employs the search performance for controlling which low-level heuristic has to be utilized for the SVM configuration generation. It is designed with the aim of optimizing the SVM multi-objective optimization problem by including the hyper meta-heuristic and Improved Particle Swarm Optimization (IPSO) algorithm. The process of SVM configuration is designed as the multi-objective optimization problem by taking the false positive (fp), false negative (fn), true positive (tp) and true negative (tn) into consideration in which the results are attained based on metrics such as precision, recall, fmeasure and accuracy, and here the model complexity is found to be the two contradicting objectives. The efficiency of the HMHIPSO-SVM framework has been assessed on two cyber security challenges, (1) malware big data classification, and then (2) anomaly intrusion detection. The novel framework shows improved classification performance of big data cyber security problems in comparison with other available algorithms.