The design and performance of intrusion detection system classifier based on the time series windows

Intruders always change their behaviors on a sudden. Due to this characteristic, we assume linearity as a characteristic of the increment of normal data stream in a fixed period of time. Based on this assumption, we obtained the increments of data stream from Agent OCX between each two successive points of time by logging them in time series windows. These data streams are delivered from a monitoring network. Subsequently, we put these increments into a Perceptran Neural Network (PNN). The adoption of this module is due to the efficient classifying quality of the PNN which enables us to distinguish the normal data from the abnormal data and helps to identify whether the networks are attacked or not. Our experiment shows that our classifier is easy to design and has a high functional quality in classifying.