Securing an Electric Vehicle Using a Predictive Model

Akshatha D · 2025 · Crossref

DOI: 10.22214/ijraset.2025.74093

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the growing cybersecurity vulnerabilities in electric vehicles (EVs) caused by increased digital connectivity and reliance on artificial intelligence. Traditional security measures often fail to detect evolving threats and lack transparency, hindering effective response. To resolve this, the authors propose an interpretable machine learning (IML) framework for intrusion detection that not only identifies cyberattacks with high accuracy but also explains its decisions to enhance trust and usability for security teams. The study utilizes the UNSW-NB15 dataset, comprising 82,332 records with 45 features representing network traffic. The methodology involves extensive preprocessing, including handling missing values, encoding categorical data, normalizing numerical features, and removing outliers using the Interquartile Range method. To address class imbalance, the Synthetic Minority Oversampling Technique (SMOTE) and class weighting were applied. Four models were trained and evaluated: XGBoost, Random Forest, K-Nearest Neighbors (KNN), and Long Short-Term Memory (LSTM). Hyperparameter tuning was performed using GridSearchCV with 5-fold cross-validation. Crucially, the framework integrates explainability tools—SHAP for global and local feature importance, Partial Dependence Plots (PDP) for feature interactions, and LIME for local instance explanations—to ensure transparency. The results demonstrate that XGBoost achieved the highest performance, recording 97.23% accuracy, 98.91% precision, 97.12% recall, and a 98.01% F1-score. It effectively balanced sensitivity and specificity, with high precision (99%) and recall (97%) for attack traffic. Random Forest followed closely with 96.86% accuracy, while LSTM achieved 95.14% accuracy, excelling in capturing temporal dependencies. KNN served as a reliable baseline with 93.55% accuracy. Interpretability analysis revealed that features such as `sttl`, `dbytes`, and `dur` significantly influenced predictions. SHAP and LIME provided clear insights into how specific feature values drove classification decisions, validating the model's logic. The significance of this work lies in its integration of robust predictive performance with explainable AI, addressing the "black-box" limitation of many intrusion detection systems. By providing transparent insights into threat detection, the framework enables EV manufacturers and cybersecurity teams to understand and act on alerts with confidence. The study concludes that this interpretable, AI-driven approach offers a scalable and resilient solution for strengthening EV cybersecurity against diverse intrusion types, including Denial-of-Service and remote access attacks.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• telematics crash prediction