Asian Journal of Research in Computer Science

Cloud computing introduces significant security challenges, particularly sophisticated intrusion attempts that threaten data integrity and service availability. Traditional intrusion detection systems (IDS) prioritize detection accuracy while overlooking computational resource constraints, making them impractical for resource-limited cloud environments. This paper proposes a novel multi-objective bio-inspired optimization framework that simultaneously maximizes intrusion detection accuracy while minimizing computational overhead through adaptive ensemble learning. The framework employs NSGA-III (Non-dominated Sorting Genetic Algorithm III) for multi-objective feature selection, balancing detection performance with resource consumption across four objectives: accuracy, feature count, processing time, and memory usage. A hybrid Whale Optimization Algorithm-Particle Swarm Optimization (WOA-PSO) approach optimizes ensemble weights for a heterogeneous ensemble of deep learning and machine learning classifiers. An adaptive three-tier framework automatically adjusts model complexity based on available computational resources across edge, fog, and cloud infrastructure layers. Experimental evaluation on the KDD Cup 1999 dataset demonstrates that the proposed framework achieves 92.42% detection accuracy while reducing feature dimensionality by 71.3% (from 122 to 35 features), significantly lowering computational requirements. The NSGA-III optimization yields Pareto-optimal solutions enabling flexible trade-offs between detection performance and resource efficiency without substantial accuracy degradation. Although evaluation is limited to the KDD Cup 1999 benchmark, the framework’s multi-objective approach provides generalizable principles for resource-aware security. The framework enables practical deployment across heterogeneous infrastructure from battery-powered IoT devices to cloud servers, with demonstrated inference latency of 0.3 ms at edge tier enabling real-time security monitoring. The adaptive deployment strategy allows seamless scaling from resource-constrained IoT sensors to high-performance cloud servers, enabling real-time intrusion detection on battery-powered edge devices while maintaining competitive accuracy. This research contributes a practical, scalable solution for deploying intrusion detection in resource-constrained cloud environments, demonstrating that bio-inspired multi-objective optimization can effectively balance security requirements with operational efficiency constraints.