Sliva, A., Neal Reilly, S., Blumstein, D., Hookway, S., and Chamberlain, J.
Presented at the 7th International Conference on Applied Human Factors and Ergonomics (AHFE 2016), Orlando, FL (July 2016)
Awarded Best Paper at AHFE’s Cross-Cultural Decision-Making Conference
When planning operations or designing policy interventions, military decision-makers and policy experts must have an understanding of the world around them, particularly the political, military, economic, social, information, and infrastructure (PMESII) implications on and effects of these policies. Unfortunately, it is notoriously difficult to rigorously analyze causality in complex sociocultural systems. Controlled experiments are often impossible and capturing the rich and complex causal and predictive dynamics of a sociocultural system is challenging: the diversity of cultural, political, social, economic, and historical factors and data sources that comprise these systems strongly suggests that such modeling cannot be successful without incorporating multiple domains of study and a variety of analytic methods.
In this paper, we present three types of ensemble combinations that can be leveraged to mitigate the practical impacts of PMESII modeling complexity: (1) using a variety of analytic approaches that work over different types of data sets to leverage all available data (e.g., combining temporal and non-temporal data); (2) applying different analytic approaches to address different aspects of the sociocultural system (e.g., one approach for identifying related features, another for determining the direction of causation); and (3) combining the results of using several parallel modeling approaches and combining their results into a single coherent analysis by applying insights from machine learning where similar types of ensembles have been consistently able to outperform individual learning approaches. There are a variety of techniques for combining modeling approaches for all of these ensemble combinations, including supervised, semi-supervised, and human-guided approaches. While many of the causal analyses combined in the ensemble will be automated, not all of them must be, and we also explore various approaches for incorporating human expertise into the ensemble mechanism.
To demonstrate the efficacy of this ensemble approach, we present results from three experiments using a mixture of synthetic and real-world data sets. First, we developed an ensemble combination of additive noise models to determine the causality from observational, non-temporal data. Second, we constructed an ensemble of diverse causal analysis techniques for time-series data that leverage metadata to improve performance. In both of these cases, the ensemble outperformed the individual approaches and produced a more accurate model of causal relationships. Third, we used a human-in-the-loop process to develop an ensemble model for extracting causal relationships from real-world data about Anbar province in Iraq, using both temporal and non-temporal data and techniques to form a more complete picture of the causal dynamics and PMESII effects in the region.
The views, opinions, and/or findings expressed are those of the author(s) and should not be interpreted as representing the official views or policies of the Department of Defense or the U.S. Government. Distribution Statement “A” (Approved for Public Release, Distribution Unlimited)
For More Information
To learn more or request a copy of a paper (if available), contact Amy Sliva.
(Please include your name, address, organization, and the paper reference. Requests without this information will not be honored.)