Platform Utilizing Similar Users' Data to Detect Anomalous Operation of Home IoT without Sharing Private Information
Platform Utilizing Similar Users' Data to Detect Anomalous Operation of Home IoT without Sharing Private Information
复制标题
平台利用相似用户的数据来检测家庭物联网的异常操作，而无需共享私人信息
DOI：
10.1109/access.2021.3112482
复制
发表时间：
2021
期刊：
影响因子：
3.9
通讯作者：
Masayuki Murata
中科院分区：
文献类型：
--
作者：
Masaaki Yamauchi;Yuichi Ohsita;Masayuki Murata
为了降低家庭物联网设备受到网络攻击的风险，我们提出了一种通过基于家庭物联网设备的操作序列和家庭条件学习用户行为来检测异常操作的方法。虽然这种方法需要足够的训练数据，但通过利用具有相似生活方式的用户的数据，仍然可以实现准确的检测。然而，用户不愿意与他人分享他们的私人信息。在这项研究中，我们提出了一个平台，利用类似用户的数据，而不共享私人信息。我们引入了一个代理，学习用户的行为，以检测异常操作在每个家庭和其他代理合作。在这个框架中，需要与其他代理合作的代理向其他代理发送问题，并附上与所学习的行为相似的过去问题的标识符。接收者通过使用附加信息来决定问题是否来自类似的代理。如果问题来自类似的代理，则代理回答问题。我们通过使用从真实的家庭收集的行为数据集来评估我们的平台。我们模拟了两种情况：（1）监控操作序列，以及（2）单独使用家庭物联网设备，但序列不能用于检测。结果表明，当使用每个用户的行为数据时，我们的框架对情况（1）的检测率高出50.5%。对于情况（2），当使用用户的所有行为数据时，我们的框架具有13.4%的高检测率。
To mitigate the risk of cyberattacks on home IoT devices, we have proposed a method for detecting anomalous operations by learning the behaviors of users based on the operation sequences of their home IoT devices and home conditions. While this method requires a sufficient amount of training data, achieving accurate detection is still possible by utilizing the data of users with similar lifestyles. However, users are unwilling to share their private information with others. In this study, we propose a platform to utilize data of similar users without sharing private information. We introduce an agent that learns behaviors of users to detect anomalous operations in each home and cooperates with other agents. In this framework, an agent requiring cooperation with other agents sends a question to the other agents, attaching identifiers of past questions that are similar to the behaviors learned. The receivers decide whether the question is from a similar agent by using the attached information. If the question is from a similar agent, the agent answers the question. We evaluate our platform by using behavior datasets collected from real homes. We simulate two cases: (1) sequences of operations are monitored, and (2) home IoT devices are used alone but sequences cannot be used for detection. The results show that our framework has a 50.5% higher detection ratio for case (1) when using the behavioral data of each user. For case (2), our framework has a 13.4% higher detection ratio when using all the behavioral data of users.