An IoT security framework using deep/machine learning techniques for smart offices

智能办公室是一种利用信息和通信技术来提高员工生产力和优化资源利用的工作场所。新兴的基于物联网（IoT）的智能办公自动化（SOA）应用包括但不限于节能、物理安全、智能采暖等。物联网应用程序从物理对象收集数据，并将其发送到中央服务器进行处理和生成命令。收集的数据和用户命令的安全性是重中之重，也是广泛采用基于物联网的技术的关键障碍。网络攻击可能导致经济损失、健康危害，甚至生命损失。拟议中的项目将把IT（信息技术）服务提供商Cistech Ltd.与滑铁卢大学的网络安全专家联系起来，开发一种基于机器/深度学习（M/D-L）的安全解决方案，以检测SOA应用程序物联网网络中的入侵。该团队将共同：1)设计一个软件框架来评估不同的M/D-L模型，这些模型使用最先进的优化技术来自动选择物联网设备和网络数据；2)设计一种基于不同M/D-L模型超参数整定和多选择标准的模型选择器。由于分类模型的准确率和虚警率对超参数的取值很敏感，因此在设计选择方法时要考虑提高准确率和降低虚警率的多重目标。设计了一种利用不同集成算法的超参数调谐来选择最佳模型的技术。选择技术将使用多个分类评估指标：检测率、准确性、f分数和误报率；3)为基于物联网的SOA应用设计概念验证型入侵检测系统。各种攻击将被模拟，从目标2中选择的模型将被用来证明所提出的入侵检测系统的有效性。加拿大拥有强大的IT生态系统，涵盖物联网设备的各个方面，包括硬件、传感器、通信、安全和M/D-L技术。该项目的成果将扩大加拿大基于物联网应用的安全技术范围。

A smart office is a work-place where information and communication technologies are used to increase employee productivity and optimize resource usage. The emerging IoT (Internet of Things) based smart office automation (SOA) applications include, but are not limited to, energy conservation, physical security, and smart heating. IoT applications collect data from physical objects and send it to a central server for processing and generating commands. The security of collected data and user's commands is a top priority, and a key barrier to widespread take-up of IoT-based technologies. A cyber-attack can lead to financial losses, health hazards, and even loss of life. The proposed project will connect Cistech Ltd., an IT (Information Technology) service provider, with cybersecurity experts at the University of Waterloo to develop a machine/deep-learning (M/D-L) based security solution to detect intrusions in IoT networks for SOA applications. Together, the team will: 1) Design a software framework to evaluate different M/D-L models that use state-of-the-art optimization techniques for automated feature selection IoT device and network data; 2) Design a model selector based on hyper-parameter tuning of different M/D-L models and multiple selection criteria. Since the accuracy and false alarm rates of classification models are sensitive to the values of hyper-parameters, selection methods will be designed with multiple objectives: increasing accuracy and minimizing false alarm rate. A technique will be designed to select the best model using hyper-parameter tuning of different ensemble algorithms. The selection technique will use multiple classification evaluation metrics: detection rate, accuracy, F-score, and false alarm rate; 3) Design a proof-of-concept intrusion detection system for IoT-based SOA applications. A variety of attacks will be simulated, and the selected model from objective 2 will be used to demonstrate the efficacy of the proposed intrusion detection system. Canada has a strong IT ecosystem covering all aspects of IoT devices, including hardware, sensors, communication, security, and M/D-L techniques. Results from this project will expand the envelope of security technologies for IoT-based applications in Canada.