Abstract: In recent years, DDoS attack detection has mostly adopted machine learning methods, and Stacking is one of them. The current stacking base-learner configuration method is mostly fixed collocation. Due to the complexity and dynamics of DDoS attacks, static configuration strategy is obviously less flexible. In this regard, the QGA-Stacking algorithm is proposed, which uses quantum genetic algorithm (QGA) to dynamically select a group of learner combinations with the highest evaluation index in Stacking, thereby improving the accuracy and flexibility of the detection model. At the same time, a set of optimal feature sets was proposed to save computational cost. Through experimental comparison, it is fully proved that the QGA-Stacking algorithm has more significant detection performance than the other three mainstream algorithms, and the selection of the best feature set is more reasonable.