A Crowding Distance-Driven Adaptive Genetic Algorithm with Multiobjective Evolutionary Optimization for Robust Attack Detection

With the growing sophistication of cyber-attacks, traditional detection mechanisms struggle to maintain high accuracy while minimizing false positives and false negatives. The selection of optimal feature subsets and configurations significantly affects detection performance, necessitating advanced optimization strategies. Conventional Genetic Algorithms (GAs) lack dynamic adaptability and often converge prematurely due to low diversity. Moreover, single-objective optimization fails to balance conflicting objectives such as maximizing detection accuracy while minimizing false alarms. This study proposes an Adaptive Genetic Algorithm (AGA) integrated with Crowding Distance Selection (CDS) in a Multiobjective Evolutionary Algorithm (MOEA) framework to optimize binary-encoded feature subsets and system configurations. The proposed AGA dynamically adjusts mutation rates based on population diversity metrics calculated using the Hamming distance. Higher mutation rates are triggered when diversity is low to encourage exploration. The crowding distance is used to select diverse solutions in the multiobjective space defined by accuracy, false positives, and false negatives. A diverse initial population of feature subsets is evolved using adaptive mutation and crossover, guided by fitness scores from MOEA. Experimental evaluation on the NSL-KDD dataset reveals that the proposed AGA-MOEA method achieves a detection accuracy of 97.4%, a false positive rate of 2.1%, and a false negative rate of 3.8%.