Affective DDoS Attack Detection in EV Charging Station Using DL Autoencoder Model

Distributed Denial of Service (DDoS) attacks are increasingly targeting advanced grid-based electric vehicle (EV) charging stations, posing serious risks to service availability and grid stability. To address these challenges, this study evaluates the effectiveness of machine learning and deep learning approaches using the CICEV2023 dataset, which contains 126,568 samples across four attack scenarios (e.g., Wrong EV, Wrong Timestamp) and multiple attack types (e.g., TCP flood, SYN flood, cryptojacking). This study compares the performance of the Light Gradient Boosting Machine (LightGBM) and a proposed Deep Neural Network (DNN) enhanced with autoencoder-based feature extraction. The results demonstrate that the DNN model significantly outperforms LightGBM, achieving 87.01% accuracy, 87.65% precision, 87.3% recall, and 88.8% F1-score. These findings highlight the potential of autoencoder-assisted deep learning for detecting sophisticated DDoS threats in EV charging infrastructures. Future work will explore scalability, computational efficiency, and integration with real-world charging networks.