{"id":{"repo_id":"brock","oai_identifier":"oai:brocku.scholaris.ca:10464/19649"},"canonical_url":"https://search.dev.ndltd.org/etd/brock/oai:brocku.scholaris.ca:10464/19649","repository":{"repo_id":"brock","name":"Brock University","base_url":"https://brocku.scholaris.ca/server/oai/request"},"display":{"title":"BERT-Based Intrusion Detection System for RF Jamming Attacks in Vehicular Network","abstract":"As vehicular networks continue to evolve toward increased connectivity and autonomy, they become more vulnerable to cybersecurity threats, particularly Radio Frequency (RF) jamming attacks that can severely disrupt communication systems. This thesis presents a comprehensive study on the application of transformer-based models for RF jamming detection in intelligent transportation systems. Specifically, we evaluate the performance of four pre-trained transformer architectures—BERT, RoBERTa, DistilBERT, and ALBERT—fine-tuned for the multi-class classification of interference, smart jamming, and constant jamming attacks under varying vehicular speeds (25 m/s and 15 m/s). These models are systematically compared against traditional machine learning baselines, including K-Nearest Neighbors (KNN), Random Forest (RaFo), and Long Short-Term Memory (LSTM) networks. Our results show that transformer models significantly outperform the baselines in terms of classification accuracy and robustness under mobility-induced signal variation. At high speed (25 m/s), BERT achieved the best performance with an accuracy of 96.25%, while at low speed (15 m/s)—a more challenging scenario due to reduced temporal dynamics—DistilBERT led with an accuracy of 91.00%. In contrast, baseline models like KNN and RaFo experienced substantial performance drops, falling to 82.27% and 80.04%, respectively. In addition to detection performance, we conducted an in-depth analysis of putational efficiency metrics, including training time, inference speed, FLOPs, and GPU memory usage. DistilBERT emerged as the most resource-efficient model, training 59% faster and using 31% less memory than BERT, while maintaining comparable accuracy. RoBERTa demonstrated strong accuracy but incurred the highest computational cost, and ALBERT, though compact in size, showed only moderate performance gains. Confusion matrix analysis revealed that all transformer models classified interfer- ence scenarios with near-perfect accuracy. However, distinguishing between smart and constant jamming attacks remained challenging, especially at lower speeds. Importantly, none of the transformer models misclassified attack traffic as benign, indicating strong reliability from a cybersecurity standpoint. This study demonstrates that transformer-based architectures offer a viable and effective approach to RF jamming detection in vehicular networks, balancing high detection performance with practical deployment efficiency. Future work will explore lightweight fine-tuning techniques and real-world deployment scenarios.","abstract_html":"As vehicular networks continue to evolve toward increased connectivity and autonomy, they become more vulnerable to cybersecurity threats, particularly Radio Frequency (RF) jamming attacks that can severely disrupt communication systems. This thesis presents a comprehensive study on the application of transformer-based models for RF jamming detection in intelligent transportation systems. Specifically, we evaluate the performance of four pre-trained transformer architectures—BERT, RoBERTa, DistilBERT, and ALBERT—fine-tuned for the multi-class classification of interference, smart jamming, and constant jamming attacks under varying vehicular speeds (25 m/s and 15 m/s). These models are systematically compared against traditional machine learning baselines, including K-Nearest Neighbors (KNN), Random Forest (RaFo), and Long Short-Term Memory (LSTM) networks. Our results show that transformer models significantly outperform the baselines in terms of classification accuracy and robustness under mobility-induced signal variation. At high speed (25 m/s), BERT achieved the best performance with an accuracy of 96.25%, while at low speed (15 m/s)—a more challenging scenario due to reduced temporal dynamics—DistilBERT led with an accuracy of 91.00%. In contrast, baseline models like KNN and RaFo experienced substantial performance drops, falling to 82.27% and 80.04%, respectively. In addition to detection performance, we conducted an in-depth analysis of putational efficiency metrics, including training time, inference speed, FLOPs, and GPU memory usage. DistilBERT emerged as the most resource-efficient model, training 59% faster and using 31% less memory than BERT, while maintaining comparable accuracy. RoBERTa demonstrated strong accuracy but incurred the highest computational cost, and ALBERT, though compact in size, showed only moderate performance gains. Confusion matrix analysis revealed that all transformer models classified interfer- ence scenarios with near-perfect accuracy. However, distinguishing between smart and constant jamming attacks remained challenging, especially at lower speeds. Importantly, none of the transformer models misclassified attack traffic as benign, indicating strong reliability from a cybersecurity standpoint. This study demonstrates that transformer-based architectures offer a viable and effective approach to RF jamming detection in vehicular networks, balancing high detection performance with practical deployment efficiency. Future work will explore lightweight fine-tuning techniques and real-world deployment scenarios.","abstract_has_math":false,"creators":["Wickramasurendra Nujitha"],"institution":"Brock University","degree_name":"M.Sc. Computer Science","degree_level":"Master","degree_discipline":"Faculty of Mathematics and Science","degree_department":"Department of Computer Science","school":null,"contributors":[],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025-09-17","date_published":"2025-09-17","updated_at":"2026-07-24T01:23:05Z","subjects":["TECHNOLOGY::Information technology::Computer science::Software engineering","TECHNOLOGY::Information technology::Computer science::Computer science"],"languages":["eng"],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://hdl.handle.net/10464/19649","outbound_label":"Handle","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.department","label":"Department","values":["Department of Computer Science"]},{"key":"dc:creator","label":"Author","values":["Wickramasurendra Nujitha"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2025-09-17T17:27:25Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2025-09-17T17:27:25Z"]},{"key":"dc:date.issued","label":"Date","values":["2025-09-17"]},{"key":"dc:publisher","label":"Institution","values":["Brock University"]},{"key":"dc:type","label":"Dc Type","values":["Electronic Thesis or Dissertation"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Faculty of Mathematics and Science"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Master"]},{"key":"thesis:degree_name","label":"Degree Name","values":["M.Sc. Computer Science"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["TECHNOLOGY::Information technology::Computer science::Software engineering","TECHNOLOGY::Information technology::Computer science::Computer science"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["eng"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/10464/19649"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["As vehicular networks continue to evolve toward increased connectivity and autonomy, they become more vulnerable to cybersecurity threats, particularly Radio Frequency (RF) jamming attacks that can severely disrupt communication systems. This thesis presents a comprehensive study on the application of transformer-based models for RF jamming detection in intelligent transportation systems. Specifically, we evaluate the performance of four pre-trained transformer architectures—BERT, RoBERTa, DistilBERT, and ALBERT—fine-tuned for the multi-class classification of interference, smart jamming, and constant jamming attacks under varying vehicular speeds (25 m/s and 15 m/s). These models are systematically compared against traditional machine learning baselines, including K-Nearest Neighbors (KNN), Random Forest (RaFo), and Long Short-Term Memory (LSTM) networks. Our results show that transformer models significantly outperform the baselines in terms of classification accuracy and robustness under mobility-induced signal variation. At high speed (25 m/s), BERT achieved the best performance with an accuracy of 96.25%, while at low speed (15 m/s)—a more challenging scenario due to reduced temporal dynamics—DistilBERT led with an accuracy of 91.00%. In contrast, baseline models like KNN and RaFo experienced substantial performance drops, falling to 82.27% and 80.04%, respectively. In addition to detection performance, we conducted an in-depth analysis of putational efficiency metrics, including training time, inference speed, FLOPs, and GPU memory usage. DistilBERT emerged as the most resource-efficient model, training 59% faster and using 31% less memory than BERT, while maintaining comparable accuracy. RoBERTa demonstrated strong accuracy but incurred the highest computational cost, and ALBERT, though compact in size, showed only moderate performance gains. Confusion matrix analysis revealed that all transformer models classified interfer- ence scenarios with near-perfect accuracy. However, distinguishing between smart and constant jamming attacks remained challenging, especially at lower speeds. Importantly, none of the transformer models misclassified attack traffic as benign, indicating strong reliability from a cybersecurity standpoint. This study demonstrates that transformer-based architectures offer a viable and effective approach to RF jamming detection in vehicular networks, balancing high detection performance with practical deployment efficiency. Future work will explore lightweight fine-tuning techniques and real-world deployment scenarios."]},{"key":"dc:title","label":"Title","values":["BERT-Based Intrusion Detection System for RF Jamming Attacks in Vehicular Network"]}]}],"canonical_facts":{"dc:contributor.department":["Department of Computer Science"],"dc:creator":["Wickramasurendra Nujitha"],"dc:date.accessioned":["2025-09-17T17:27:25Z"],"dc:date.available":["2025-09-17T17:27:25Z"],"dc:date.issued":["2025-09-17"],"dc:description.abstract":["As vehicular networks continue to evolve toward increased connectivity and autonomy, they become more vulnerable to cybersecurity threats, particularly Radio Frequency (RF) jamming attacks that can severely disrupt communication systems. This thesis presents a comprehensive study on the application of transformer-based models for RF jamming detection in intelligent transportation systems. Specifically, we evaluate the performance of four pre-trained transformer architectures—BERT, RoBERTa, DistilBERT, and ALBERT—fine-tuned for the multi-class classification of interference, smart jamming, and constant jamming attacks under varying vehicular speeds (25 m/s and 15 m/s). These models are systematically compared against traditional machine learning baselines, including K-Nearest Neighbors (KNN), Random Forest (RaFo), and Long Short-Term Memory (LSTM) networks. Our results show that transformer models significantly outperform the baselines in terms of classification accuracy and robustness under mobility-induced signal variation. At high speed (25 m/s), BERT achieved the best performance with an accuracy of 96.25%, while at low speed (15 m/s)—a more challenging scenario due to reduced temporal dynamics—DistilBERT led with an accuracy of 91.00%. In contrast, baseline models like KNN and RaFo experienced substantial performance drops, falling to 82.27% and 80.04%, respectively. In addition to detection performance, we conducted an in-depth analysis of putational efficiency metrics, including training time, inference speed, FLOPs, and GPU memory usage. DistilBERT emerged as the most resource-efficient model, training 59% faster and using 31% less memory than BERT, while maintaining comparable accuracy. RoBERTa demonstrated strong accuracy but incurred the highest computational cost, and ALBERT, though compact in size, showed only moderate performance gains. Confusion matrix analysis revealed that all transformer models classified interfer- ence scenarios with near-perfect accuracy. However, distinguishing between smart and constant jamming attacks remained challenging, especially at lower speeds. Importantly, none of the transformer models misclassified attack traffic as benign, indicating strong reliability from a cybersecurity standpoint. This study demonstrates that transformer-based architectures offer a viable and effective approach to RF jamming detection in vehicular networks, balancing high detection performance with practical deployment efficiency. Future work will explore lightweight fine-tuning techniques and real-world deployment scenarios."],"dc:identifier.uri":["https://hdl.handle.net/10464/19649"],"dc:language.iso":["eng"],"dc:publisher":["Brock University"],"dc:subject":["TECHNOLOGY::Information technology::Computer science::Software engineering","TECHNOLOGY::Information technology::Computer science::Computer science"],"dc:title":["BERT-Based Intrusion Detection System for RF Jamming Attacks in Vehicular Network"],"dc:type":["Electronic Thesis or Dissertation"],"thesis:degree_discipline":["Faculty of Mathematics and Science"],"thesis:degree_level":["Master"],"thesis:degree_name":["M.Sc. Computer Science"]},"updated_at":"2026-07-24T01:23:05Z"}