Jayasri C, Balaji V, Nalayini C M, Pradeep S
Department of Electronics and Communication Engineering, AVC College of Engineering, Mayiladuthurai, 609305, Tamil Nadu, India.
Department of Electronics and Communication Engineering, Easwari Engineering College, Ramapuram, Chennai, 600089, Tamil Nadu, India.
Sci Rep. 2025 May 31;15(1):19141. doi: 10.1038/s41598-025-04643-8.
The growing adoption of intelligent transportation systems and connected vehicle networks has raised significant cybersecurity concerns due to their vulnerability to cyberattacks such as spoofing, message tampering, and denial-of-service. Traditional intrusion detection systems struggle to cope with the dynamic and high-volume nature of vehicular data, often leading to high false positives and limited adaptability. To address this problem, this study proposes an enhanced deep learning-based optimization framework for detecting cyberattacks in vehicle networks. The methodology employs the UNSW-NB15 dataset, with data preprocessed using Maximum-Minimum Normalization. Feature extraction is performed using the Discrete Fourier Transform (DFT), capturing frequency-domain patterns indicative of anomalies. Detection is executed through an Improved Long Short-Term Memory (ILSTM) model, whose parameters are optimized using the Crocodile Optimization Algorithm (COA), aiming to maximize classification accuracy. Experimental results demonstrate that the proposed ILSTM-COA model significantly outperforms existing techniques, achieving 98.9% accuracy and showing notable improvements across sensitivity, specificity, and other performance metrics. This model offers a robust, scalable, and real-time solution for safeguarding vehicular networks against evolving cyber threats.
智能交通系统和车联网的日益普及引发了重大的网络安全问题,因为它们容易受到诸如欺骗、消息篡改和拒绝服务等网络攻击。传统的入侵检测系统难以应对车辆数据的动态性和高容量特性,常常导致误报率高且适应性有限。为了解决这个问题,本研究提出了一种基于深度学习的增强优化框架,用于检测车辆网络中的网络攻击。该方法采用了UNSW-NB15数据集,并使用最大-最小归一化对数据进行预处理。使用离散傅里叶变换(DFT)进行特征提取,捕捉表示异常的频域模式。通过改进的长短期记忆(ILSTM)模型进行检测,其参数使用鳄鱼优化算法(COA)进行优化,旨在最大化分类准确率。实验结果表明,所提出的ILSTM-COA模型显著优于现有技术,准确率达到98.9%,并且在灵敏度、特异性和其他性能指标方面都有显著提高。该模型为保护车辆网络免受不断演变的网络威胁提供了一个强大、可扩展的实时解决方案。