Abstract:
In real world applications, it is quite common for datasets to contain missing data due to a variety of limitations. A handful of techniques have been developed to addres...Show MoreMetadata
Abstract:
In real world applications, it is quite common for datasets to contain missing data due to a variety of limitations. A handful of techniques have been developed to address this problem and impute the missing intervals. The majority of the developed techniques have targeted missing completely at random (MCAR) and missing at random (MAR) datasets and none of them gives a measure of uncertainty. In this paper, the issue of missing data imputation in time series analysis is addressed from a different angle where special attention is devoted to not missing at random (NMAR) datasets and the associated uncertainty characterization. For this purpose, Kriging type techniques as well as Bayesian Updating (BU), commonly used in spatial statistics, are applied and the results are compared to those of more standard techniques. The outcomes of this comparison show the superiority of the adaptedtechniques both in improving predictability and providing the possibility of uncertainty quantification.
Published in: Proceedings of the 2014 IEEE 15th International Conference on Information Reuse and Integration (IEEE IRI 2014)
Date of Conference: 13-15 August 2014
Date Added to IEEE Xplore: 02 March 2015
Electronic ISBN:978-1-4799-5880-1