Date of Award


Document Type



Santa Clara : Santa Clara University, 2023.


Computer Science and Engineering

First Advisor

Behnam Dezfouli


Smart home devices are becoming increasingly popular across the United States. Unfortunately, in order to keep the cost of these devices down, manufacturers tend to place cybersecurity measures lower on their list of priorities. As a result, these devices have become targets for security breaches like the Mirai Botnet and distributed denial of service (DDoS) attacks. The cybersecurity solutions currently available on the market are either expensive, difficult to set up, or follow a static set of rules for attack detection. In this thesis, we propose a novel approach to detecting attacks on Internet of Things (IoT) devices through the use of a forward proxy powered by a machine learning model running on a Linux machine. The proposed solution is more affordable, requires minimal configuration, and dynamically adjusts detection rules based on prior attacks. The use of a machine learning model in the solution allows us to take advantage of the regularity in IoT traffic patterns to identify and isolate potentially malicious packets. The model was trained using the IoT-23 data set, which provides a labeled set of malicious and benign packets from IoT devices. We verified the effectiveness of the approach using a testbed of various smart home devices set up in an isolated home network. These devices range in price and usage context, allowing us to ensure that the solution will be effective with the majority of smart home devices. After testing, we have determined that the model is able to successfully identify and isolate of malicious packets sent from a compromised IoT device if the device matches a device that was used in training data. Moving forward, we hope to implement the solution directly onto the router hardware using the P4 language for widespread adoption.

Available for download on Thursday, July 11, 2024