Fast-flux networks is a domain name system (DNS) technique used by botnets, which is hiding some attack like phishing and malware delivery sites behind associate dynamical network of compromised hosts acting as proxies, that sometimes hosts malicious content. Detection of fast-flux networks continues to be a difficult issue attributable to the similar behavior between these networks and alternative legitimate infrastructures, like server farms and content distribution networks. This study seeks to improve the detection and prediction of the unknown “zero-day” online fast-flux botnet. This improvement will be achieved using a new system called the fast-flux hunter (FFH), which supports a new adaptive evolving fuzzy neural network algorithm. The FFH system is a hybrid between the supervised and unsupervised online knowledge-based learning systems. The core mechanism of the FFH is based on the inherent feature of the fast-flux networks. It uses a collection of DNS traffic information. The FFH is able to scan over 7615 domain records and extract 14 distinct features for each domain. The FFH decreases the classification method’s error rate. The FFH has a detection accuracy rate of approximately 98 % and is compatible with life-long learning systems, footprint-consuming memory, and high-speed systems.