Modeling Information Blackouts in Missing Not-At-Random Time Series

Résumé fourni par la source

This release provides the first stable, fully reproducible codebase accompanying our preprint on modeling traffic-sensor "blackouts" as Missing Not At Random (MNAR) in a latent state-space framework. Highlights MNAR latent state-space model with missingness mechanism (EM-trained) for blackout-aware imputation and forecasting Baselines for comparison (e.g., LOCF / interpolation + seasonal naive / MAR LDS) and evaluation scripts End-to-end pipeline: data loading/preprocessing → window generation → training → evaluation → figure generation Reproducible configs + scripts to regenerate main results and plots What's included Core model + training code (Kalman filter/smoother + EM) Experiment runners and evaluation utilities (imputation + multi-horizon forecast) Plotting scripts used to generate the paper figures How to reproduce Clone the repo Follow the README Run the provided scripts/configs to regenerate the primary tables/figures Related links Preprint: https://arxiv.org/abs/2601.01480 Project repo: https://github.com/BlackoutBayes/Modeling-Information-Blackouts-in-MNAR-Time-Series Zenodo This release is intended for Zenodo archival + DOI minting via GitHub→Zenodo integration.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.