摘要
用户击键行为作为一种生物特征,具有采集成本低、安全性高的特点。然而,现有的研究方法和实验环境都是基于实验室数据,并不适用于极度不平衡的真实数据。比如,在实验室数据上效果出色的分类算法在真实数据上却无法应用。针对此问题,提出了基于真实击键行为数据的用户识别算法。该方法将聚类算法和距离算法结合起来,通过比较新来的击键行为和历史击键行为相似度以实现用户识别。实验结果表明,该算法在100名用户的3 015条真实击键记录组成的数据集上准确率达到88.22%,在投入实际应用后,随着样本集的增大算法的准确率还可以进一步提升。
Keystroke dynamics, part of biometrics, is featured as low cost and high security. However, existing researches and experiments are mostly based on laboratory data, which are not appropriate for extremely unbalanced real data. For example, classification algorithms are not applicable to real data due to the extreme imbalance of normal and abnormal samples. In order to solve this problem, a keystroke dynamics method was proposed for real data, which combined classification algorithm with distance algorithm. It authenticated users by comparing user” s new behavior with historical behavior. Experimental results show that this method has an accuracy rate of 88. 21% on the real dataset of 3 015 records of 100 users. It is expectable that the performance will be better with the expansion of data set.
出处
《计算机应用》
CSCD
北大核心
2015年第A01期110-112,129,共4页
journal of Computer Applications
基金
国家自然科学基金重大研究计划项目(91218301)
国家自然科学基金青年项目(60903201)
中央高校基本科研业务专项项目(JBK140129)
关键词
键盘行为
用户识别
欧氏距离
K-MEANS聚类
生物认证
keystroke dynamics
user authentication
Euclidean distance
k-means clustering
biometric authentication
