基于动态特征选择的Android应用隐私风险自动化检测  

AUTOMATED DETECTION OF PRIVACY RISKS IN ANDROID APPLICATIONS BASED ON DYNAMIC FEATURE SELECTION

作　　者：高龙良 杜素果[1] 杨金萍 Gao Longliang;Du Suguo;Yang Jinping(Antai College of Economics and Managements,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]上海交通大学安泰经济与管理学院,上海200240

出　　处：《计算机应用与软件》2024年第6期312-319,共8页Computer Applications and Software

基　　金：国家自然科学基金项目(71671114)。

摘　　要：针对Android应用中可能存在的用户隐私泄露问题,提出一种基于机器学习方法的自动化检测模型。该模型选择使用App申请的权限作为特征,动态地选取特征集,并采用四种经典的机器学习算法进行独立的训练与预测,最终确定最适用于Android应用的隐私风险检测模型。实验结果表明,对于隐私风险应用,该模型能够实现平均95%以上的识别准确率。该模型能够从多层面更好地进行应用风险管理以及用户隐私保护,具有较高的社会效益与实际应用价值。Aimed at the user privacy leakage problem that may exist in Android applications,an automated detection model based on machine learning methods is proposed.This model chose to use the permission items applied by the App as features,dynamically selected the feature set,and used four classical machine learning algorithms to independently train and predict.And the most suitable privacy risk detection model for Android applications was determined.Experimental results show that the model can achieve an average prediction accuracy of more than 95% for privacy risk applications.This model can better manage application risk and protect user privacy from multiple aspects,which has high social benefit and practical value.

关 键 词：隐私风险 权限 机器学习 动态特征选择 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]