TinyML-Enabled Resource-Efficient Framework for Real-Time Network Securiy in Electric Vehicle Charging Networks
Bibliographic record
Abstract
Electric Vehicle Charging Infrastructure (EVCI) faces critical cybersecurity challenges due to its integration with external networks, making it vulnerable to attacks such as malware propagation, session hijacking, and Denial-of-Service (DoS). Traditional machine learning (ML) models for intrusion detection are computationally intensive, making them unsuitable for resource-constrained Electric Vehicles (EVs). This paper presents a novel multi-stage framework that enables real-time intrusion detection through TinyML models. The proposed framework combines advanced feature selection (FS) using SHapley Additive exPlanations (SHAP) values with hybrid pruning techniques to create efficient, deployment-ready models. Using the CICEVSE2024 dataset, the paper demonstrates significant improvements across multiple architectures: Multi-Layer Perceptron (MLP) achieves 95.9% reduction in inference time and 80.7% reduction in model size while maintaining more than 94% accuracy; Long Short-Term Memory (LSTM) shows 69% reduction in inference time and 79.2% size reduction with more than 94% accuracy; and Extreme Gradient Boosting (XGBoost) delivers 15.7% reduction in inference time and 91.2% reduction in size while preserving more than 98% accuracy. We further evaluate the models on a Raspberry Pi 5 (8 GB) and observe that although inference times increase on the resource-constrained device, the pruned models consistently maintain real-time performance, using the same models as on the desktop to ensure identical accuracies for a fair comparison. These results establish the feasibility of deploying TinyML-based intrusion detection systems (IDSs) directly on EVs, enabling robust real-time cybersecurity without exceeding hardware constraints.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.004 | 0.000 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.011 |
| Science and technology studies | 0.002 | 0.000 |
| Scholarly communication | 0.001 | 0.000 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".