Incomplete information imputation in limited data environments with application to disaster response

Kezban Yagci Sokat*, Irina S. Dolinskaya, Karen Smilowitz, Ryan Bank

*Corresponding author for this work

Research output: Contribution to journalArticle

4 Scopus citations

Abstract

Following a major disaster, a field operations manager needs to deploy relief activities within the affected region. State-of-the-art humanitarian logistics models have been developed over the past decades to assist relief operations. However, while many models assume availability of information on infrastructure status, this is typically not the case in practice. Often, only partial information about infrastructure status is known. Utilizing the similarities in the known data via attributes, we develop a framework to impute incomplete information in limited data environments. We present an application of this framework to a past disaster, the 2010 Haiti earthquake. We build an ArcGIS model to automate the data collection and processing efforts to the extent possible. The study explores the impact of missing data, dispersion of missing data and imputation techniques used in approximating the incomplete information. Our results suggest that lower granularity yields better estimates of the unknown information above a threshold. We also develop publicly available test cases for the broader community.

Original languageEnglish (US)
Pages (from-to)466-485
Number of pages20
JournalEuropean Journal of Operational Research
Volume269
Issue number2
DOIs
StatePublished - Sep 1 2018

    Fingerprint

Keywords

  • 2010 Haiti earthquake
  • Humanitarian logistics
  • Imputation
  • Incomplete information
  • Limited data

ASJC Scopus subject areas

  • Computer Science(all)
  • Modeling and Simulation
  • Management Science and Operations Research
  • Information Systems and Management

Cite this