Adaptive Control charts for identifying concept drift in nonstationary environment

Time adjusting dynamic systems whose underlying changing distribution should be continuously
monitored to track abnormal behaviors is one of the most recent challenges in many real life
applications. In fields such as sensor networks, intrusion detection, credit card fraud detection and
process monitoring, the arriving data change over time and the target concept to be learned changes
accordingly causing the problem of “concept drift. Adaptive control charts from SPC domain and
dynamic ensemble methods from data mining field are the most widely used techniques to track
concept drift. In order to perform the change identification in data stream processes, the first
enhancement of ensemble methods in SPC proposed in Mejri et al, 2018 entitled Dynamic Weighted
Majority Control Chart will be presented. The new adaptive method has not only the ability to
combine more than two control charts but also uses the expertise of dynamic weighted majorityWinnow (DWM)-WIN in
identifying and learning changes during the monitoring. First, it transforms
the task of determining the state of the process into a classification problem by treating control charts
as attributes where the drift has to be predicted. Second, DWM-WIN mechanism is applied to learn the
shift and to combine the decision of different individuals. Third, a prediction of class label is used to
help in classifying the shift during the changing of the process toward the approximated right
direction. The three main steps of this combined control chart as well comparative results will be
presented and discussed in this talk