Hybrid Models for Effective Adversarial Attack Detection in Cyberspace Using Machine Learning
Bhavesh Kumar Sharma, Atul Kumar Rai, Pawan Kumar, Asim Kumar Rai, Khushboo Tripathi · 2025
This With the exponential proliferation of cyberspace, it brought us in to a new area of security challenges that require cutting-edge methods and tools for adversaries to launch attack. The paper describes a novel idea of hybrid model using latest machine learning algorithms in order to handle adversarial attacks on Internet. The hybrid architecture combines a mix of algorithms that cheats the accuracy and ensures robustness against new cyber threats. It achieves latest performance with 97.8% detection rate on the benchmark datasets via extensive analysis and experiments, outperforming standalone models by a significant margin (+12.3 %). On targeted perturbations too, the hybrid approach exhibits 93.5% detection accuracy and proves to be more computationally secure against adversaries with diverse adversarial attacks (vs. standard model). This research should be a framework to measure the performance of hybrid architectures generally and especially for enhancing security in cyberspace using advanced machine learning algorithms.