A system and method for detecting abnormal traffic behavior. The method comprises: applying a task to an input data set to create an un-normalized cluster of traffic features, wherein the task defines a plurality of traffic features; computing a center point of the cluster of traffic features; computing a distance between the computed center point and a new sample, wherein the new sample includes traffic features defined in the task; and determining, based on the computed distance, whether the received new sample demonstrates abnormal behavior.