In this paper, we propose a novel black-box approach to adversarial attacks on structured data such as network traffic. To overcome the dependency on system access rights and repetitive exploration of previous studies, we present a method to minimize interactions for detection evasion and real-world scenario reflection. Sensitive features are identified and perturbed through an adaptive feature selection strategy using changepoint detection and causal analysis. The lightweight design results in low computational cost and easy deployment, and experiments demonstrate detection evasion, adaptability, and practical applicability.
