John Kelly’s research while affiliated with North Carolina Agricultural and Technical State University and other places

What is this page?


This page lists works of an author who doesn't have a ResearchGate profile or hasn't added the works to their profile yet. It is automatically generated from public (personal) data to further our legitimate goal of comprehensive and accurate scientific recordkeeping. If you are this author and want this page removed, please let us know.

Publications (6)


Fig. 1: Federated Learning Architecture for vehicular networks. CAV clients share model updates (or parameters) from previous training to a centralized server for aggregation into a new update for continuous training.
Fig. 2: The FedFT framework: Server-side pre-training on proxy data, and client-side fine-tuning on localized data for IDS classification and inference.
Fig. 3: IDS model architecture: Pre-trained CNN Module with Conv1D-MaxPooling layers for feature extraction, and a Multilayer Perceptron module with Fully Connected and ReLU layers for fine-tuning.
Fine-Tuning Federated Learning-Based Intrusion Detection Systems for Transportation IoT
  • Preprint
  • File available

February 2025

·

2 Reads

·

Nana Kankam Brym Gyimah

·

Mansi Bhavsar

·

John Kelly

The rapid advancement of machine learning (ML) and on-device computing has revolutionized various industries, including transportation, through the development of Connected and Autonomous Vehicles (CAVs) and Intelligent Transportation Systems (ITS). These technologies improve traffic management and vehicle safety, but also introduce significant security and privacy concerns, such as cyberattacks and data breaches. Traditional Intrusion Detection Systems (IDS) are increasingly inadequate in detecting modern threats, leading to the adoption of ML-based IDS solutions. Federated Learning (FL) has emerged as a promising method for enabling the decentralized training of IDS models on distributed edge devices without sharing sensitive data. However, deploying FL-based IDS in CAV networks poses unique challenges, including limited computational and memory resources on edge devices, competing demands from critical applications such as navigation and safety systems, and the need to scale across diverse hardware and connectivity conditions. To address these issues, we propose a hybrid server-edge FL framework that offloads pre-training to a central server while enabling lightweight fine-tuning on edge devices. This approach reduces memory usage by up to 42%, decreases training times by up to 75%, and achieves competitive IDS accuracy of up to 99.2%. Scalability analyses further demonstrates minimal performance degradation as the number of clients increase, highlighting the framework's feasibility for CAV networks and other IoT applications.

Download

FL-IDS: Federated Learning-Based Intrusion Detection System Using Edge Devices for Transportation IoT

January 2024

·

567 Reads

·

15 Citations

IEEE Access

·

·

·

[...]

·

A federated learning-based intrusion detection system (FL-IDS) is introduced in this paper to enhance the security of vehicular networks in the context of IoT edge device implementations. The FL-IDS system protects data privacy by using local learning, where devices share only model updates with an aggregation server. This server then generates an enhanced detection model. The FL-IDS system also incorporates machine learning (ML) and deep learning (DL) classifiers, namely logistic regression (LR) and convolutional neural networks (CNN), to prevent attacks in transportation IoT environments. The performance of the proposed IDS was evaluated using two different datasets, NSL-KDD and Car-Hacking. The model evaluation has been evaluated based on the accuracy and loss parameters. The results showthat the FL-IDS system outperforms traditional centralized machine learning and deep learning approaches regarding accuracy and privacy protection.


Anomaly-based intrusion detection system for IoT application

May 2023

·

568 Reads

·

74 Citations

Discover Internet of Things

Internet-of-Things (IoT) connects various physical objects through the Internet and it has a wide application, such as in transportation, military, healthcare, agriculture, and many more. Those applications are increasingly popular because they address real-time problems. In contrast, the use of transmission and communication protocols has raised serious security concerns for IoT devices, and traditional methods such as signature and rule-based methods are inefficient for securing these devices. Hence, identifying network traffic behavior and mitigating cyber attacks are important in IoT to provide guaranteed network security. Therefore, we develop an Intrusion Detection System (IDS) based on a deep learning model called Pearson-Correlation Coefficient - Convolutional Neural Networks (PCC-CNN) to detect network anomalies. The PCC-CNN model combines the important features obtained from the linear-based extractions followed by the Convolutional Neural Network. It performs a binary classification for anomaly detection and also a multiclass classification for various types of attacks. The model is evaluated on three publicly available datasets: NSL-KDD, CICIDS-2017, and IOTID20. We first train and test five different (Logistic Regression, Linear Discriminant Analysis, K Nearest Neighbour, Classification and Regression Tree,& Support Vector Machine) PCC-based Machine Learning models to evaluate the model performance. We achieve the best similar accuracy from the KNN and CART model of 98%, 99%, and 98%, respectively, on the three datasets. On the other hand, we achieve a promising performance with a better detection accuracy of 99.89% and with a low misclassification rate of 0.001 with our proposed PCC-CNN model. The integrated model is promising, with a misclassification rate (or False alarm rate) of 0.02, 0.02, and 0.00 with Binary and Multiclass intrusion detection classifiers. Finally, we compare and discuss our PCC-CNN model in comparison to five traditional PCC-ML models. Our proposed Deep Learning (DL)-based IDS outperforms traditional methods.


Anomaly-based Intrusion Detection System for IoT Application

April 2023

·

222 Reads

·

1 Citation

Internet-of-Things (IoT) connects various physical objects through the Internet and it has a wide application, such as in transportation, military, healthcare, agriculture, and many more. Those applications are increasingly popular because they address real-time problems. In contrast, the use of transmission and communication protocols have raised serious security concerns for IoT devices, and traditional methods such as signature and rule-based methods are inefficient for securing these devices. Hence, identifying network traffic behavior and mitigating cyber attacks are important in IoT to provide guaranteed network security. Therefore, we develop an Intrusion Detection System (IDS) based on a deep learning model called Pearson-Correlation Coefficient - Convolutional Neural Networks (PCC-CNN) to detect network anomalies. The PCC-CNN model combines the important features obtained from the linear-based extractions followed by the Convolutional Neural Network. It performs a Binary classification for Anomaly detection and also a Multiclass classification for various types of attacks. The model is evaluated on three publicly available datasets: NSL-KDD, CICIDS-2017, and IOTID20. We first train and test five different (Logistic Regression, Linear Discriminant Analysis, K Nearest Neighbour, Classification and Regression Tree,\& Support Vector Machine) PCC-based Machine Learning models to evaluate the model performance. We achieve the best similar accuracy from the KNN and CART model of 98\%, 99\%, and 98\%, respectively, on the three datasets. On the other hand, we achieve a promising performance with a better detection accuracy of 99.89\% and with a low misclassification rate of 0.001 with our proposed PCC-CNN model. The integrated model is promising, with a misclassification rate (or False alarm rate) of 0.02, 0.02, and 0.00 with Binary and Multiclass intrusion detection classifiers. Finally, we compare and discuss our PCC-CNN model in comparison to five traditional PCC-ML models. Our proposed Deep Learning (DL)-based IDS outperforms traditional methods.


A Survey of QEMU-based Fault Injection Tools & Techniques for Emulating Physical Faults

January 2023

·

265 Reads

·

3 Citations

IEEE Access

Fault Injection (FI) is a method used to quantify the reliability and resilience of a system by assessing the system’s ability to detect, locate, and/or mitigate fault occurrences. At the architecture level, targeted bit flips at specific times and locations can help quantify the response of a running application to unwanted changes in state and memory values. FI campaigns of this type can be performed on the target hardware virtual implementations of the target device. In this paper, we present a survey of Quick EMUlator (QEMU) based FI techniques. After discussing the various techniques proposed by academia and industry, we classified them into categories and compare their attributes. This review will help researchers understand the capabilities and limitations of using the QEMU emulator for FI-based system reliability analysis. Additionally, we identify the gaps in existing techniques and propose opportunities for extensions.


Figure 5 train data box plot for model comparison of binary and multiclass classification
Intrusion-Based Attack Detection Using Machine Learning Techniques for Connected Autonomous Vehicle

August 2022

·

87 Reads

·

5 Citations

Lecture Notes in Computer Science

With advancements in technology, an important issue is ensuring the security of self-driving cars. Unfortunately, hackers have been developing increasingly complex and harmful cyberattacks, making them difficult to detect. Furthermore, due to the diversity of the data exchanged amongst these vehicles, traditional algorithms face difficulty detecting such threats. Therefore, a network intrusion detection system is essential in a connected autonomous vehicle's communication infrastructure. The IDS (intrusion detection system) aims to secure the network by identifying malicious and abnormal traffic in real-time. This paper focuses on the data preprocessing, feature extraction, attack detection for such a system.Additionally, it will compare the performance of this proposed IDS when operating in different machine learning models. We apply Linear Regression (LR), Linear Discriminant Analysis (LDA), K Nearest Neighbors (KNN), Classification and Regression Tree (CART), and Support Vector Machine (SVM) to classify the NSL-KDD dataset. The dataset was classified using binary and multiclass classification to train and test files. This data resulted in 94% and 98% accuracy for the train and test files, respectively, with KNN and CART algorithms.KeywordsMachine learningAutonomous vehicleCyberattacksIntrusionData preprocessingFeature engineeringML modelAccuracy

Citations (3)


... This scheme presents high detection accuracy and low false positive rates however, it lacks scalability. A federated learning-based IDS, FL-IDS, using edge devices is proposed in [66]. In this approach, logistic regression and convolution neural network classifiers are trained at the vehicular level and the locally trained models are aggregated by the edge server. ...

Reference:

A Survey on Distributed Approaches for Security Enhancement in Vehicular Ad-hoc Networks
FL-IDS: Federated Learning-Based Intrusion Detection System Using Edge Devices for Transportation IoT

IEEE Access

... Another direction in the literature uses PCC to measure the correlation between the features and targeted class. Bhavsar et al. (Bhavsar et al. 2023) proposed a model of lightweight botnet detection in IoT. They adopted PCC feature selection and Convolutional Neural Network (CNN) as learning classifier. ...

Anomaly-based intrusion detection system for IoT application

Discover Internet of Things

... The study by [13] presents an approach to fortify selfdriving cars against complex cyber threats. Their focus involves preprocessing data, feature extraction, and attack detection using various machine learning models such as logistic regression (LR), linear discriminant analysis(LDA), K-nearest neighbors(KNN), classification and regression trees (CART), SVM on the NSL-KDD dataset, attaining notable accuracies of 94% and 98% for training and testing datasets, respectively, stands out, particularly when employing KNN and CART algorithms. ...

Intrusion-Based Attack Detection Using Machine Learning Techniques for Connected Autonomous Vehicle

Lecture Notes in Computer Science