FSD-LAB/noc-incident-dataset-for-ml

NOC Incident Dataset for Machine Learning

This dataset contains preprocessed and feature-engineered samples for network device incident prediction in a Network Operations Center (NOC) setting. The raw operational logs and telemetry have already been cleaned, aggregated, transformed, and vectorized for direct use in classical machine learning workflows.

Data Format

The dataset is distributed as a SciPy CSR sparse matrix:

• File: ml_data_vec.npz
• Format: compressed SciPy sparse matrix (scipy.sparse.save_npz)
• Shape: (581906, 62979)
• Sparse format: CSR

Each row corresponds to one sample. The feature columns contain vectorized log-derived and engineered telemetry features. The final two columns have fixed meanings:

• Second-to-last column: Unix timestamp for the sample
• Last column: binary incident label

Loading Example

from scipy import sparse

matrix = sparse.load_npz("ml_data_vec.npz")

X = matrix[:, :-2]
timestamps = matrix[:, -2].toarray().ravel()
y = matrix[:, -1].toarray().ravel().astype(int)

Intended Use

This release is intended for evaluating machine learning models for short-horizon network incident prediction, especially feature-engineered tabular approaches such as LightGBM. The dataset is already vectorized, so users can train directly on X without rerunning the original preprocessing pipeline.

Notes

• The timestamp column is provided to support chronological splitting and time-aware evaluation.
• The label column indicates whether the corresponding sample is associated with an incident target.
• The original raw logs and intermediate preprocessing artifacts are not included in this vectorized release.