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

Le résumé fourni par la source

This preprint studies traffic-sensor blackouts (contiguous missing intervals) and evaluates two tasks: (1) imputation, i.e., reconstructing values inside blackout windows, and (2) post-blackout forecasting at horizons +1, +3, and +6 steps on a 5-minute grid. We compare a MAR linear dynamical system (Kalman filtering with RTS smoothing) against an MNAR extension that treats the missingness mask as an informative observation channel using a logistic missingness model conditioned on the latent state. The repository includes code, evaluation-window manifests, and notebooks for experiments on the Seattle Loop dataset and the METR-LA dataset.

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