Unveiling Backdoor Attacks in Networking with Machine Learning Approaches
Anshika Sharma, Himanshi Babbar · 2024
The maintenance of cybersecurity and the protection of sensitive information depend on detecting Backdoor attacks in network environments. This study uses the UNSW-NB15 dataset, a well-known network intrusion detection systems (NIDS) benchmark, and machine learning (ML) techniques to build strong models that can detect Backdoor attacks. To facilitate thorough analysis and model training, the dataset offers various internet traffic samples, including benign and malicious cases. The goal is to enable reliable and timely detection of Backdoor attacks by identifying trends and abnormalities through feature design and model optimization. By comparing how well several ML algorithms including Random Forest (RF), Naive Bayes (NB), Extreme Gradient Boosting (XGBoost), and K-Nearest Neighbour (KNN) detect backdoor intrusions, the study helps develop IDS. Compare the performance of the models and find the best method for identifying backdoor attacks by looking at criteria like accuracy, precision, recall, and F1-score. With useful insights for enterprises trying to reduce the risks of backdoor attacks and fortify their defences against ever-changing cyber threats, this study's results can be applied to improve cybersecurity in network environments. The best ML models for identifying backdoor attacks are the focus of this research. Also, it might be useful for deciding which detection strategies to use and how to use them in an actual IoT environment. The XGBoost model outperforms the others with an accuracy rate of 95.61%, on scale with KNN at 90.52%, RF at 89.56%, and NB at 74.21%.