The continuous advancement in the Internet of Things technology allows people to connect anywhere at any time, thus showing great potential in technology like smart devices (including smartphones and wearable devices). However, there is a possible risk of unauthorized access to these devices and technologies. Unfortunately, frequently used authentication schemes for protecting smart devices (such as passwords, PINs, and pattern locks) are vulnerable to many attacks. USB tokens and hardware keys have a risk of being lost. Biometric verification schemes are insecure as well as they are susceptible to spoofing attacks. Maturity in sensor chips and machine learning algorithms provides a better solution for authentication problems based on behavioral biometrics, which aims to identify the behavioral traits that a user possesses, such as hand movements and waving patterns. Therefore, this research study aims to provide a solution for passive and continuous authentication of smartphone users by analyzing their activity patterns when interacting with their phones. The motivation is to learn the physical interactions of a smartphone owner for distinguishing him/her from other users to avoid any unauthorized access to the device. Extensive experiments were conducted to test the performance of the proposed scheme using random forests, support vector machine, and Bayes net. The best average recognition accuracy of 74.97% is achieved with the random forests classifier, which shows the significance of recognizing smartphone users based on their interaction with the phones.

物联网技术的不断进步使人们可以随时随地进行连接，从而在智能设备（包括智能手机和可穿戴设备）等技术方面显示出巨大的潜力。然而，这些设备和技术可能存在未经授权访问的风险。不幸的是，用于保护智能设备的常用身份验证方案（例如密码、PIN 和模式锁）很容易受到许多攻击。 USB 令牌和硬件密钥有丢失的风险。生物特征验证方案不安全，并且容易受到欺骗攻击。传感器芯片和机器学习算法的成熟为基于行为生物识别技术的身份验证问题提供了更好的解决方案，旨在识别用户拥有的行为特征，例如手部动作和挥手模式。因此，本研究旨在通过分析智能手机用户与手机交互时的活动模式，为智能手机用户的被动和持续身份验证提供解决方案。其动机是学习智能手机所有者的物理交互，以将他/她与其他用户区分开来，以避免对设备的任何未经授权的访问。使用随机森林、支持向量机和贝叶斯网络进行了大量的实验来测试所提出方案的性能。使用随机森林分类器实现了 74.97% 的最佳平均识别准确率，这显示了根据智能手机用户与手机的交互来识别智能手机用户的重要性。