APP下载

基于RBF神经网络的Android恶意行为识别

2018-08-06陈天伟

现代电子技术 2018年15期

陈天伟

摘 要: 传统Android恶意行为识别方法无法解决恶意行为特征的动态波动性,识别出的恶意行为精度差,并且需要耗费大量的时间,因此提出基于RBF神经网络的Android恶意行为识别方法。该方法首先进行Android恶意行为的样本采集、行为特征提取和数据整合,使输出的结果可以被RBF神经网络识别,然后采用RBF神经网络局部逼近的特点提高学习速度,增强神经网络结果的输出质量,并采用K均值聚类算法得到所有特征集中各样本到该特征集中心距离的平方和,取该距离的最小值,通过最小二乘递推法计算隐含层节点到数据输出层节点的权值,完成RBF神经网络的训练,实现Android恶意行为的准确识别。实验结果说明所提方法可以提高对Android恶意行为识别的正确率和效率。

关键词: RBF神经网络; Android恶意行为; 识别; 特征集; 局部逼近; 权值

中图分类号: TN915.08?34; TP391 文献标识码: A 文章编号: 1004?373X(2018)15?0083?04

Android malicious behavior recognition based on RBF neural network

CHEN Tianwei1, 2

(1. University of Electronic Science and Technology of China, Chengdu 610101, China; 2. Urban Vocational College of Sichuan, Chengdu 610110, China)

Abstract: The traditional Android malicious behavior recognition method can′t solve the dynamic fluctuations of malicious behavior characteristics, has poor accuracy of malicious behavior recognition, and spends a large amount of time. Therefore, the Android malicious behavior recognition method based on RBF neural network is proposed. The items of sample acquisition of Android malicious behavior, behavior feature extraction and data integration are carried out to make that the output result can be identified by RBF neural network. The local approximation characteristic of RBF neural network is adopted to improve the output quality of neural network and learning speed. The K?means clustering algorithm is adopted to get the quadratic sum of distances from each sample in the feature set to the center of the feature set, so as to obtain the minimum distance. The least square recursive method is used to calculate the weight from the node of hidden layer to the node of data output layer, accomplish the training of RBF neural network, realize the accurate identification of Android malicious behavior. The experimental results show that the proposed method can improve the accuracy and efficiency of Android malicious behavior recognition.

Keywords: RBF neural network; Android malicious behavior; recognition; feature set; local approximation; weight

0 引 言

隨着经济的快速发展,移动互联网应用领域日益扩大,智能Android系统也已成为市场占有量最大的操作系统之一,面对日益庞大的Android数据库,采用有效的行为识别方法对Android恶意和正常的行为进行识别成为目前急需解决的问题。传统的LSSVM Android恶意行为识别方法采用LSSVM基于恶意行为特征进行识别,无法解决恶意行为特征的动态波动性,识别出的恶意行为精度差,并且需要耗费大量的时间[1],为解决该问题,本文提出基于RBF神经网络的Android恶意行为识别方法,提高对Android恶意行为识别的正确率和效率,得到了满意度较高的识别结果。……

登录APP查看全文