Securing the Internet of Things with Deep Neural Networks
This study provides an objective solution to defending networks of Internet of Things (IoT) devices. There is currently no comprehensive solution to defending the IoT, let alone a protocol for IoT security. Recent attacks compromised over 120 million devices high lighting this lack of effective security.
This thesis investigates the effectiveness of deep learning for critical security applications of IoT devices by utilizing snapshots of network traffic from nine real-world devices. This differentiates from other research methodologies since IoT researchers typically settle for simulated network traffic. Support Vector Machines (SVM), Random Forest and Deep Neural Network (DNN) are tested and compared against one another to determine which is the most deployable and provide the highest accuracy of anomaly detection. The results are assisting to deliver a scalable and deployable software solution for commercial use.
All of the algorithms scored high accuracies. The Deep Neural Network provides the highest coefficient of determination compared to the other tested models, implying that this is the best model on the tested data. Finally, the DNN's learning autonomy omits humans from the loop resulting in an optimum real-world algorithm.