Research On Database User Abnormal Behavior Detection Based On Semi-supervised Learning

With the development of science and technology,the role of the Internet in the world is growing,data security is the most important part of Internet security.Data stored in the database,the security of the database is the same as the data security is threatened,once the database data leakage then the loss is difficult to estimate.The research shows that the threat of internal attack to database security is much greater than that of external attack,so it is very important to detect the abnormal behavior of legitimate users in database.Based on the work of predecessors,this paper proposes a semi-supervised learning-based method to detect the abnormal behavior of database users.This paper presents a new method to describe the behavior of database users in the form of five tuples and extract their features.Five-tuple extraction of database user behavior features has integrated the advantages of existing methods,can be a complete and concise description of the database user behavior features.Tri-training algorithm is a semi-supervised learning algorithm,which uses three basic classifiers to train the training data set on the basis of a single view,and the number of training iterations is too many.An improved Tri-training Algorithm is proposed,in which the idea of ensemble learning is used in the base classifier and the concept of bifurcation rate is introduced to control the training times to reduce the generation of noisy data,it reduces the possibility that iterative training leads to reverse optimization and reduces the training time.The database user's abnormal behavior detection system based on the improved Tri-training Algorithm is tested and the experimental results are analyzed,it is found that the improved Tri-training algorithm has obvious advantages over other algorithms in detection effect and training time.