Missing sensor network data

Abstract

We describe details of a missing data method for longitudinal motion sensor data. The method uses Markov Chain Monte Carlo computations on a Bayesian Markovian hierarchical model that is designed for computational efficiency on large data sets. The model can handle special structure of missing sensor observations including intervals of downtime, corrupted neighboring data, and highly correlated observations.

Keywords:

• Bayesian hierarchical model
• Markov Chain Monte Carlo
• missing data
• motion sensor data

Additional information

Funding

This work was supported under NIH (P30-AG024978). Data were provided by ORCATECH at Oregon Health and Science University with funding provided by NIH (R01-AG024059, R01-AG042191, and P30-AG008017).