Microserviços para geração de RCLS no GRASP-FS: Uma abordagem escalável para seleção de atributos em IDSs industriais

Abstract

This paper presents a scalable microservices-oriented architecture, called Distributed RCL Generator (DRG), to decouple and parallelize the construction phase in the Greedy Randomized Adaptive Search Procedure for Feature Selection (GRASP-FS) metaheuristic in Feature Selection (FS) for intrusion detection. In GRASP-FS, the construction phase generates initial feature subsets that are optimized in the local search phase. As a proof-of-concept based on the Kafka framework and four FS algorithms, we demonstrated that the adopted algorithm strategy impacts the intrusion detection F1-Score in a Cyber-Physical Systems scenario ranging from 50.47% to 84.92%. Finally, we show that parallel processing can accelerate the construction phase about 3.4 times.