Hongwei Zhang

Conducted network security research under the guidance of the supervisor, with the subject direction of the theory and application of wireless sensor networks and blockchain.
Innovatively proposed a distributed ledger network topology and published in SCI journal.
Assisted my supervisor in reviewing academic papers related to my research direction, and provided professional comments on research methodology and experimental design. This experience not only strengthened my academic insight but also deepened my understanding of the peer review process. My feedback helped to improve the quality of the thesis and contributed to the development of the academic community.
Mentoring and evaluating the topic selection and research process of undergraduate students in the cluster, giving advice on the direction of the research and the research plan.

Tutoring undergraduate and graduate students in specialized courses, responsible for preparing, teaching, answering questions and grading assignments in laboratory and tutorial courses. This included courses such as Data Structures and Algorithms, Network Computing, and Network Security. In the process, I consolidated my professional knowledge and improved my language organization and presentation skills to efficiently and accurately express code optimization methods, code ideas, etc. to others.

Intrusion detection in Internet of Medical Things (IoMT) devices presents unique challenges due to diverse communication protocols and evolving security threats. This study proposes a hybrid deep learning approach combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks for enhanced intrusion detection in healthcare IoMT devices. Our CNN-LSTM model leverages CNN for spatial feature extraction and LSTM for temporal pattern analysis, particularly suited for IoMT-generated data patterns. Using the recently proposed CICIoMT2024 dataset, which covers three primary IoMT communication protocols (Bluetooth, WiFi, and MQTT), we conducted comprehensive evaluations across 18 different attack types categorized into five major classes: DDoS, DoS, Reconnaissance, MQTT, and Spoofing. The experimental results demonstrate the model's excellent performance, achieving 86.24% accuracy in multi-classification tasks and 99.79% accuracy in binary classification scenarios. The model maintains consistently high performance across other metrics, with precision, recall, and F1-scores of 0.865, 0.863, and 0.863 respectively for multi-classification, and 0.998 across all metrics for binary classification.

Repository: https://github.com/Hongwei-Z/CNN-LSTM-IoMT-IDS

A framework named SC-MLIDS is proposed, which uses a server-client hybrid machine learning model for dual-layer intrusion detection in Wireless Sensor Networks (WSNs). By aggregating predictions from two models, the SC-MLIDS framework not only effectively reduces redundancy in data transmission but also achieves efficient and accurate intrusion detection, providing reliable protection for network security.

Repository: https://github.com/Hongwei-Z/SC-MLIDS

Independently conceptualized and implemented a project to integrate traditional machine learning algorithms with federated learning frameworks, using random forests as the core algorithm. Created a scenario with three clients and one server to effectively demonstrate the communication between global and local models in federated learning, achieving model aggregation through parameter exchange. Overcame the limitations in the existing capabilities of the open-source federated learning library Flower by developing a unique solution that marks a contribution to the open-source community.

Repository: https://github.com/Hongwei-Z/Federated-Random-Forest

An innovative approach is proposed to address the rapid growth of the Internet of Things (IoT) devices leading to a significant increase in their security threats: deploying Federated Learning on Software-Defined Networks (SDNs) for identifying malicious traffic from IoT devices. The goal is to reduce the security threat of botnet-infected IoT devices while protecting data privacy. Attack simulation experiments are conducted for model detection, and the results demonstrate that the model exhibits superior detection rates under four different types of attacks, with F1 scores exceeding 85% for all of them. The project received an "A+" grade for proposing a feasible solution to address the fast-growing IoT threat problem. I was mainly responsible for processing the dataset, testing the model, and summarizing and presenting the experimental results.

Repository: https://github.com/Hongwei-Z/SDN_FL_IoT_DDoS

This project is based on Canadian immigration statistics and aims to study the factors affecting immigration in Canada. KMeans, hierarchical clustering, decision trees and association rule mining are used to analyze the socio-economic conditions of the country of origin of immigrants and Canada, and the reasons why immigrants leave their home countries and the conditions that make Canada attractive to immigrants are examined separately. The results of the analysis show that the decision-making factors of immigrants vary for each of the seven indicators. For example, immigrants from 41 countries were influenced by the growth rate of GDP per capita in their home country. In contrast, the attractiveness factors for Canada were employment, government health expenditure, GDP per capita, etc. The project was completed individually with a final grade of "A".

Repository: https://github.com/Hongwei-Z/CanadianImmigrationStudy

This project proposed a new blockchain topology for Wireless Sensor Networks (WSNs) called Fishing Net Topology (FNT). Like Tangle, each node needs to be approved by the other two nodes. The difference is that this topology is simpler and more efficient than Tangle, the selection of Tips does not need to wait for a long time, and each new node has a defined location and defined Tips. FNT is more suitable for WSNs than Tangle.

Repository: https://github.com/Hongwei-Z/FishingNetTopology

Wireless Sensor Networks (WSNs) are limited by the shortcomings that make their nodes vulnerable to attacks. This project proposed a WSN node trust model based on IOTA Tangle and blockchain technology. The security of the WSN is significantly enhanced by allowing the data generated by two neighbouring sensors to validate the data generated by the current sensor in the Tangle. The project received an "A" grade.

Designed and completed the Community Garden Scheduler application, which provides users with integrated functions such as publishing, accepting, monitoring and reminding garden tasks. I was mainly responsible for the implementation of UI design, weather API access, and task list retrieval and filtering functions. Through this project, I gained a comprehensive understanding of the entire software development process, such as requirements analysis, design, development, testing and iteration. The project received an "A+" grade.

Repository: https://github.com/Hongwei-Z/CommunityGardenScheduler