摘要
为提高日常行为识别准确度的同时使应用具有更强的便捷性,提出基于智能手机中4种无需许可传感器数据对5种日常行为进行识别的方法。在分析Android系统传感器框架的基础上开发集成了一个小型应用程序进行数据采集处理,然后利用机器学习算法实现手机用户的行为特征识别,目标是实现个人行为准确及时且长期有效的动态监督或预测。实验结果表明,改进马尔可夫链算法与SVM分类器结合使用结果最优,测试识别准确率可接近95%,精确度、召回率等其他指标均呈现很好的效果。
To improve the accuracy and make it more convenient in the use of human behavior identification at the same time,a method using the date of four no-permission-imposed sensors in Android smartphone to recognize five kinds of daily activities was proposed.After analyzing the framework of sensors in Android system,an application was integrated for data collection and processing.Then machine learning algorithms were used to extract the features for activity recognition,which aimed at dynamic supervision or predict individual behavior accurately,timely,chronically and effectively.The result shows that the combined use of the improved Markov chain algorithm and SVM classifier have the best result,and the accuracy is close to95%,the accuracy,recall rate and other indicators are also very good.
作者
郭渊博
孔菁
刘春辉
王一丰
GUO Yuanbo;KONG Jing;LIU Chunhui;WANG Yifeng(Cryptography Engineering Institute, Information Engineering University, Zhengzhou 450001, China)
出处
《通信学报》
EI
CSCD
北大核心
2018年第A02期164-172,共9页
Journal on Communications
基金
国家自然科学基金资助项目(No.61602515
No.61501515)~~