Nsll Kdd Hdc Eval

nsll-kdd-hdc-eval

Intrusion Detection in IoT Networks Using Hyperdimensional Computing: A Case Study on the NSL-KDD Dataset — Ghazal Ghajari et al. (2025) (arXiv:2503.03037, 2025)

What this evaluates

Evaluates the ability of a hyperdimensional computing framework to detect and classify network intrusions in IoT environments. It probes the model's capacity to encode high-dimensional feature vectors, learn class prototypes, and accurately distinguish between normal traffic and specific attack types (DoS, probe, R2L, U2R).

Datasets

• NSL-KDD — total ?; splits: train (-1), test (-1)

Metrics

• accuracy (primary) — range: percent
  • The proportion of correctly classified samples out of the total number of test samples, calculated as (correct predictions / total predictions) * 100. Reported as a percentage.

Input / output format

Input: A network traffic sample represented as a vector of numerical features. Each feature is mapped to a discrete bin, and the entire sample is transformed into a 10,000-dimensional binary hypervector via XOR-based encoding and binarization.

Output: A single class label from the set {Normal, DoS, Probe, R2L, U2R}, assigned by selecting the class representative with the highest cosine similarity to the sample's hypervector.

Scoring recipe

def compute_accuracy(predictions, gold_labels):
    correct = sum(1 for p, g in zip(predictions, gold_labels) if p == g)
    return (correct / len(gold_labels)) * 100

Common pitfalls

• The binarization threshold T is typically set to half the number of features, but sensitivity to this threshold can significantly alter hypervector sparsity and classification boundaries.
• Iterative training updates (50 iterations with learning rate alpha) are applied to class representatives; stopping early or choosing an inappropriate alpha can lead to suboptimal convergence and lower accuracy.
• Cosine similarity is computed on binarized hypervectors rather than raw continuous vectors, which changes the geometric interpretation of similarity compared to standard HDC implementations.

Evidence (verbatim from paper)

In comparison to these methods, the proposed approach using hyperdimensional computing-based classification achieved a remarkable 99.5% accuracy on the NSL-KDD dataset, as shown in Table I.

Citation

@misc{ghajari2025intrusion,
  title={Intrusion Detection in IoT Networks Using Hyperdimensional Computing: A Case Study on the NSL-KDD Dataset},
  author={Ghazal Ghajari et al. (2025)},
  year={2025},
  note={arXiv:2503.03037}
}

• arXiv: 2503.03037