基于支持向量机的击键辅助认证系统

发布时间：2018-06-02 01:48

本文选题：击键认证 + 支持向量机　；参考：《西南财经大学》2014年硕士论文

【摘要】：随着社会的发展,科技进步带来的便利性越来越多体现在人们的生活中,例如电子商务将传统商业理念和先进的网络技术的结合,完全颠覆了人们的生活方式,人们可以足不出户完成交易。在进行网上交易给我们带来便利的同时,仅仅凭靠账户和密码完成的网上交易身份认证显然满足不了现在的人们对网络安全需求,因此身份识别技术正在受到越来越多的关注。传统的身份识别技术安全性可靠程度较弱,于是可靠程度较高的生物特征识别技术渐渐进入我们的视线。击键认证作为生物特征识别领域的重要一员,在近年来一直被广泛研究。传统的击键认证使用的都是击键的持续时间以及按键偏移时间,然而在前人大量的实验结果看来,这样简单的提取击键生物特征是不足以描述用户的击键行为的。另外,用来验证实验模型真实有效性的样本数据是否能反应用户在真实场景下行为,还有待考究,这也就是在击键认证领域被研究了这么多后,依旧鲜有商业化产品问世的原因之一。本文主要根据某公司的业务需求,通过对前人击键认证算法的改进,实现了完整的击键辅助认证系统。文章的主要内容有： (1)根据软件设计的规范,完成了击键认证系统架构和接口设计,另外,本系统一共实现了三种击键识别的算法,其中主要是其中的SVM支持向量机算法与另外两种算法做对比；系统流程设计与实现,系统主要包括了数据采集,数据校验与预处理,提取特征,训练模型,模型预测,模型存储等功能模块。数据上传协议设计,本文的数据采集量比较庞大,涉及面较广,为了保护用户自身隐私不受到侵犯,需要设计一种专用密码传输协议；两种基本击键模型算法的实现,为了验证本文提出的改进算法,实现了贝叶斯分类以及Manhattan距离算法,用于检验改进算法模型的性能。 (2)重点论述了支持向量机算法的基本原理以及推导过程,算法中涉及到的基本概念以及算法运行的完成流程,并对实际出现的工程问题进行了规避和防控。其中通过实验淘汰了前人实验中认为比较理想的按键特征,而根据具体实验结果选择了差异更为显著的时间特征。通过优化支持向量机内部的数据处理问题,提高了模型生成效率与模型预测效率。借助网格搜索的办法,优化模型参数,达到了自动优化惩罚系数C和高斯核函数g的效果。通过实验结果,分析用户样本的特性,如密码长度,登陆频率,登陆时间等与用户账户风险之间关系,得出结果密码长度适中,能够让真实用户不受影响,且能够很好地防范入侵者的攻击。 (3)比较改进算法和另外两种算法的优劣和不足,并提出后续改进思路。在本文系统内部测试阶段,收集到了大量数据,用户完全在无感知的状态下完成了击键认证实验,能够体现出用户的击键行为特征,实验数据真实有效。通过以这些真实数据为基础,将数据集分成训练集和测试集,用于构建模型以及检测模型算法的效果。在相同的训练集训练模型,测试集测试模型之后,Manhattan距离算法测试集正样本的预测准确率为82.87%,负样本预测准确率为81.67%；贝叶斯分类算法测试集正样本的预测准确率为87.78%,负样本的预测准确率为86.10%；支持向量机的测试集正样本的预测准确率为89.94%,负样本的预测准确率为95.80%。实验结果表明,本文的支持向量机算法,效果较为理想,达到了理想的识别精度。
[Abstract]:With the development of society , the convenience brought by scientific and technological progress is more and more reflected in people ' s life , such as the combination of traditional business concept and advanced network technology , which is not enough to describe the people ' s key behavior .

Based on the business needs of a company , this paper has realized the complete keystroke - assisted certification system by improving the key authentication algorithm of the former . The main contents of this paper are as follows :

( 1 ) According to the specification of software design , the structure and interface design of keystroke authentication system are completed . In addition , the system realizes three key recognition algorithms , in which SVM support vector machine algorithm is compared with the other two algorithms .
The system flow is designed and realized , the system mainly includes the function modules such as data collection , data checking and preprocessing , extracting features , training model , model prediction , model storage and so on . The data upload protocol is designed . The data acquisition in this paper is relatively large and the penetration is wide . In order to protect the privacy of the user , it is necessary to design a special cipher transmission protocol .
In order to validate the improved algorithm proposed in this paper , the Bayesian classification and the Manhattan distance algorithm are implemented to verify the performance of the improved algorithm model .

( 2 ) The basic principle of the support vector machine algorithm and the derivation process , the basic concept of the algorithm and the completion flow of the algorithm operation are discussed , and the time characteristic of the model generating efficiency and the model prediction efficiency is selected according to the experimental results .

( 3 ) Comparing the advantages and disadvantages of the improved algorithm and the other two algorithms , and putting forward the idea of improving the follow - up .

By taking these real data as the base , the data set is divided into training set and test set , which is used to construct the model and test model algorithm . After the same training set training model and the test set test model , the Manhattan distance algorithm test set positive sample is 82.87 % and the negative sample prediction accuracy is 81.67 % .
The prediction accuracy of positive samples was 87.78 % , and the accuracy of negative samples was 86.10 % .
The prediction accuracy rate of the test set positive samples of the support vector machine is 89.94 % , and the prediction accuracy of the negative samples is 95.80 % . The experimental results show that the support vector machine algorithm is ideal and the ideal recognition accuracy is achieved .
【学位授予单位】：西南财经大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.08

【参考文献】