Network intrusion detection is a commonly used and critical defense mechanism in the field of cybersecurity for identifying abnormal traffic online. However, the phenomenon of concept drift leads to a decrease in the accuracy of online intrusion detection systems in recognizing malicious traffic. Traditional machine learning-based intrusion detection systems are unable to adapt to the changes in data distribution of dynamic data streams. To address this issue, we propose DWOIDS, an online intrusion detection system based on dual adaptive windows and a Hoeffding tree classifier. When concept drift occurs in network data streams, it employs dual adaptive windows to monitor the prediction error of the classifier, continuously refining the classifier’s accuracy in identifying malicious traffic. We conducted experimental evaluations on multiple datasets. Our proposed method demonstrated superior classification performance when compared to the state-of-the-art.