Anomaly detection boundary based on the moving averages of Markov chain model

In the anomaly event detection and recognition, we want to know the deviation which is caused by the difference between the training Markov chain model's distribution and the real data's distribution. The moving relative entropy density deviation method is introduced to solve the problem. The results show the boundaries of the detection. If the results' fluctuations do not exceed the upper and lower boundaries, those data are normal. Otherwise, those data are dangerous.