Taylor Clark, FCLC 2025

Major: Computer Science

Bio: Taylor is a senior in the Honors program at Fordham University majoring in Computer Science and minoring in Cybersecurity and French. She is also in the accelerated program for her MS in Data Science. She discovered her passion for machine learning when she took her first data mining class last year. Since then, she has completed machine learning projects from predicting the outcome of Formula 1 races to this most recent project detecting cyber-attacks on IoMT devices, which she also presented at the IEEE Undergraduate Research Technology Conference 2024 at the Massachusetts Institute of Technology. Taylor is eager to pursue a career in machine learning, data science, or software engineering. Outside of school, she enjoys playing tennis, cross stitching, and dancing ballet.

Title of Research: Machine Learning-Based Detection for Cyber Attacks in Internet of Medical Things Devices

Mentor: Mohamed Rahouti, Ph.D., Computer and Information Sciences (CISC)

Abstract: Internet of Medical Things (IoMT) devices can connect to and use healthcare IT systems and networks, such as wearable heart rate and oxygen monitors, infusion pumps, and smart thermometers. IoMT enhances healthcare by enabling real-time patient monitoring, improving early detection and diagnosis, medication management, and reducing the need for in-person visits. IoMT also saves the healthcare industry $300 billion annually, according to Goldman Sachs. IoMT presents risks, however, particularly in data privacy and security. The U.S. healthcare system faces an annual burden of over $20 billion due to cyber attacks. Additionally, a 2022 Ponemon Institute report found that 24% of cyber-attacked hospitals noted a subsequent rise in their mortality rates. Currently, the impact of cyber attacks on healthcare is under-addressed. Accurate detection of such attacks would allow cybersecurity professionals to quickly react, saving money, sensitive information, and even lives. This project leverages machine learning (ML) to do just that. The dataset used from the Canadian Institute for Cybersecurity includes network traffic data from benign interactions and Denial of Service (DoS) and ARP Spoofing attacks on Bluetooth, WiFi and MQTT-enabled IoMT devices. Decision Tree, Random Forest, Gradient Boosting, XGBoost, Recurrent Neural Network, and Isolation Forest models were trained for each scenario. The results demonstrated the Decision Tree model performed best in detecting DoS attacks (87% accuracy), while the Random Forest model, with a max_depth of 10 and a min_samples_split of 5, performed best for ARP Spoofing detection (97% accuracy). The superior performance for ARP Spoofing detection is attributed to a more extensive feature set, allowing more precise model fitting. Overall, this study proves the efficacy of ML models for accurately detecting network attacks on IoMT devices, and underscores the need for ongoing refinement of ML techniques to address the dynamic and evolving nature of cyber threats in IoMT environments.