Malicious URL Recognition Based On Machine Learning

With the development of the Internet,the problems related to network security are becoming more and more serious.In this paper,around malicious URL recognition,several effective methods are proposed and implemented,and a malicious URL recognition system is designed.In malicious URL recognition,both black-and-white lists and rule matching have defects.Black-and-white lists cannot identify URLs that are not recorded in database,and rule matching is difficult to maintain when the number of rules is large.Traditional machine learning methods consume a lot of energy and time in feature extraction and experiment.Therefore,this paper proposes a method of neural network.It can omit the tedious step of feature engineering,and the model can evolve from low-level features to high-level features.Experiments show that the effect of neural network model is better than that of traditional detection model,and it has application value in the field of security.Firstly,a CNN model is implemented,and the results show that the effect of the model is much better than that of the traditional model.In addition,considering the lack of high-quality data sets with labels in some security areas and the fact that malicious samples often account for only a few of the total,this paper also proposes a convolutional auto-encode model to identify anomalous samples through unsupervised learning.This paper also separately detects the anomalies of domain names.For some reasons,illegal personnel will use DGA to automatically generate domain names.Therefore,if the domain name of a link is generated by the algorithm,the probability of abnormal links will be higher.Then this paper constructs a Wide & Deep network model,which can process unstructured and structured data at the same time.Finally,this paper designs a malicious sample detection system,which integrates multiple models,and these models together realize the detection of malicious samples.