Scholarlink

Electric vehicle (EV) charging infrastructures are increasingly networked, making them susceptible to cyber-physical attacks that can disrupt service and damage assets. This paper proposes a lightweight, host-based intrusion detection system (IDS) using a depth-wise separable convolutional neural network (CNN) for multiclass classification of anomalies in EV charging systems. We utilize the CICEVSE2024 dataset, applying Mutual Information (MI) for feature selection to identify the top 40 most relevant host-level kernel events. The proposed model is assessed with accuracy, precision, recall, and F1-score. Results demonstrate that the depth-wise separable CNN model achieves an F1-score of 0.9077 and an Accuracy of 0.9092 for 17-class classification, proving its effectiveness in identifying specific attack types using a reduced feature set.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Lightweight Multiclass Intrusion Detection on Electric Vehicle Infrastructure

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Mehmet Bozdal,
  • Ali Özkahraman,
  • Alper Savaşçı,
  • Zoya Pourmirza

摘要

Electric vehicle (EV) charging infrastructures are increasingly networked, making them susceptible to cyber-physical attacks that can disrupt service and damage assets. This paper proposes a lightweight, host-based intrusion detection system (IDS) using a depth-wise separable convolutional neural network (CNN) for multiclass classification of anomalies in EV charging systems. We utilize the CICEVSE2024 dataset, applying Mutual Information (MI) for feature selection to identify the top 40 most relevant host-level kernel events. The proposed model is assessed with accuracy, precision, recall, and F1-score. Results demonstrate that the depth-wise separable CNN model achieves an F1-score of 0.9077 and an Accuracy of 0.9092 for 17-class classification, proving its effectiveness in identifying specific attack types using a reduced feature set.