Anomaly Behaviors Detection Method for Online Learning Based on Copula Function

How to achieve effective anomaly behaviors detection in online learning has become a hot topic in online learning user behavior research. Traditional machine learning methods have several problems, such as poor data quality, huge amount of data, and difficulty in obtaining labels. To solve these problems, an anomaly behaviors detection method based on copula function for online learning was proposed. It estimated the tail probability distribution by using empirical copula probability. The algorithm was a non-parametric model. It decomposed the probability distribution into several independent marginal distributions. The computational complexity of the high-dimensional probability distribution could be reduced, and the computational efficiency could be effectively improved. The AUC on the public data set reached 0.91, showing that the method proposed could realize anomaly behaviors detection effectively.