基于CNN和多分类器的恶意URLs检测

Malicious URLs detection based on CNN and multiple classifiers

基于特征提取的恶意URLs的检测方法中,人工提取规则的设计依赖于丰富的专家知识以及大量的数据分析,同时规则的设计与更新需要消耗大量的人力与时间。针对以上问题,提出一种基于卷积神经网络的URLs特征自动提取方法。通过数据预处理与模型训练,实现对URLs特征的自动学习,完成提取规则的自动设计与更新。通过收敛的模型完成URLs特征的自动提取,结合J48、随机森林、支持向量机等多种分类方法对提取结果进行验证。实验结果表明,训练的模型可以自动设计和更新特征提取规则,提取的特征具有良好的区分能力及普适性。多分类器的平均准确率超过了97%,最高达到了99.2%,FPR低至0.01。

For malicious URLs detection methods based on features extraction,manual design rules depend on abundant expert knowledge and large amount of data analysis.Meanwhile designing and updating rules may cost a lot of manpower and time.To solve the problems,a URLs features extraction method based on convolutional neural network was proposed.Automatic features learning of URLs and extraction rules designing and updating were achieved by preprocessing data and training model.URLs features were extracted automatically using trained model and the features extracted were evaluated using models including J48,random forest,SVM and other classifiers.The results show that,the trained model can design and update features extraction rules automatically and the features extracted show excellent distinguishability and universality.The average precision of all classifiers reaches 97%,the maximum precision is 99.2% and FPR is as low as 0.01.