Research On Abnormal Behavior Detection Based On One-Class Classification Algorithm

Abnormal behavior refers to behavior that is significantly different from other behaviors.Although its number is small,it has great research value in many fields.The study of abnormal behavior can not only optimize the existing problems,but also find potential problems,which provides a good foundation for subsequent analysis and decision-making.But at the same time,abnormal behavior will also bring negative effects.If abnormal behavior is not handled before the experiment,it may cause serious consequences.Therefore,it has practical significance to study the method of detecting abnormal behavior.Abnormal behavior detection plays an important role in finance,communication,medical,manufacturing and other industries.Although there are many abnormal behavior detection methods,it is often difficult to obtain abnormal data in practical problems.The one-class classification algorithm is mainly based on the positive samples to establish the model,find the decision boundary,and can identify the positive and negative samples,which is suitable for the practical problems that are difficult to obtain abnormal data.Therefore,the research of abnormal behavior detection based on oneclass classification algorithm has practical significance.This thesis mainly studies abnormal behavior detection based on one-class classification algorithm.Firstly,the definition of abnormal behavior,causes and classification of abnormal behavior are given,and then a variety of methods of abnormal behavior detection are described in detail,and then the evaluation index of abnormal behavior detection and the application of abnormal behavior detection in various fields are introduced.In order to verify the abnormal behavior detection effect of one-class classification algorithm,this thesis first studies the origin and progress of one-class classification algorithm,and then focuses on the principles of three common one-class classification algorithms: One-Class Support Vector Machine,Support Vector Domain Description and Isolation Forest.After that,we propose two problems:mushroom toxicity anomaly detection and lymph state anomaly detection.We apply the innovative method based on Sparse Principal Component Analysis and Forgetting Algorithm proposed in this thesis to filter features.We use One-Class Support Vector Machine and Isolation Forest algorithm to build models for preprocessed data,and judge the effect of abnormal behavior detection and generalization performance of classifier.Experiments show that both algorithms can detect poisonous mushroom and abnormal lymph samples,and have good anomaly detection effect and good generalization performance.Therefore,abnormal behavior detection based on one-class classification algorithm has practical application value and research prospects.