摘要
针对单一k近邻算法(KNN)和最小二乘支持向量机(LSSVM)存在的缺陷,提出一种基于KNN-LSSVM的Android恶意行为识别模型.先采集Android用户行为样本,并提取相应特征组成特征向量;再将训练集输入LSSVM中进行学习,计算测试样本与最优分类平面间的距离,如果该距离小于阈值,则直接采用LSSVM恶意行为识别,否则采用KNN算法进行恶意行为识别;最后采用仿真实验测试KNN-LSSVM的性能.实验结果表明,相对于单一KNN算法和LSSVM,KNN-LSSVM提高了Android恶意行为的识别正确率,可以满足Android恶意行为的在线识别要求.
In order to solve the problem of single k nearest neighbor algorithm（KNN）and least squares support vector machine（LSSVM）and improve the identification correct rate of Android malicious behavior,the author proposed an identification model of Android malicious behavior based on KNN-LSSVM.Firstly,Android behavior samples were collected and the corresponding feature vector was extracted.Then the training samples were input into LSSVM to learn and calculate the distance between sample and classification plane.If the distance was less than threshold,LSSVM was used to recognize the malicious behavior,otherwise KNN algorithm was used to recognize the malicious behavior.Finally,the performance of KNN-LSSVM was tested by simulation experiment.The experimental results show that compared with the single KNN algorithm and LSSVM,KNN-LSSVM has improved the identification correct rate of Android malicious behavior,and can satisfy the on-line identification requirements of Android malicious behavior.
出处
《吉林大学学报（理学版）》
CAS
CSCD
北大核心
2015年第4期720-724,共5页
Journal of Jilin University:Science Edition
基金
广东省教育厅科研项目(批准号:201325531)
关键词
恶意行为
移动终端
最小二乘支持向量机
K近邻算法
malicious behavior
smart phone
least squares support vector machine
k nearest neighbor algorithm
