Project might predict serious conflicts, wars weeks in advance

Project might predict serious conflicts, wars weeks in advance
…………………………………………………………………

BY B.J. ALMOND
Rice News Staff

By mixing political
science with computer science, researchers at Rice hope
to predict serious conflicts and wars between nations four
to eight weeks in advance, with the potential to impact
foreign policy.

Rice computer
scientist Devika Subramanian and political scientist Richard
Stoll have been working together for several years using
ideas, data and theories from both their disciplines to
address this problem.

Their research
project entails using the most recent advances in computing
facilities, including a supercomputer that can perform a
trillion calculations per second, and takes advantage of
the vast expansion of computer networking to compile information
about political events in various countries over a lengthy
period of time. The data will be analyzed to identify patterns
that can help forecast militarized international conflict.

“Our ultimate
goal is to significantly improve our ability to predict
serious international conflicts,” said Stoll, professor
of political science. “We want to both develop new
techniques and adapt existing ones to create the extensive
sets of data about events between countries and to apply
models of international conflict to predict the outbreak
of military action.”

The proliferation
of news in electronic form has made such an ambitious goal
possible, he said, citing worldwide news sources that can
be accessed online, such as Reuters, Associated Press, United
Press International, cnn.com and the New York Times. Advances
in technology that can rapidly search these databases on
the World Wide Web and “mine” them for the relevant
information have also made the study feasible, Stoll added.

Previously, people
had to be trained to read through media sources to extract
and code the information by hand, a slow process that was
very time-consuming and expensive.
The researchers at Rice plan to build computer programs
that gather a large set of current and archived electronic
information sources. Then they will use techniques that
are already available to extract data about events between
countries from them.
Event data consists of an action, such as a military strike
or threat, the country that initiated the action, the country
that was the target and the date of occurrence. The actions
are scored on a scale that indicates how cooperative or
hostile the country was that initiated the action. The extracted
information will be coded so that it can be analyzed in
a variety of fashions, employing techniques from both computer
science and political science.

Subramanian,
associate professor of computer science, will apply and
extend existing algorithms for machine learning and signal
processing to analyze the events data and search for patterns
that would predict the outbreak of serious conflict. For
example, this might involve reviewing the events involving
two neighboring countries in the Middle East over two-week
periods and analyze whether dramatic changes in the pattern
of events were associated with outbreaks of conflict.

One of the key
issues is how to aggregate, or group, the data so it can
be analyzed effectively. “We’re interested in
patterns and relationships through time,” Stoll said.
The researchers want to develop new conflict-prediction
techniques that correlate event data streams across time
and geographic regions. They also want to develop models
that can track the evolution of conflict over time.

“We seek
to predict, with a lead time of four to eight weeks, the
outbreak of serious conflicts, even though they might not
reach the level of war,” Stoll said. Analyzing why
the conflict occurred will help the researchers develop
models for predicting conflict.
The researchers are well aware that event data sets can
become quite large. They estimate that a global data set
spanning the time period of the Cold War is likely to encompass
some 200 million events.

Because of the
large volume of data required for this project, the researchers
will take advantage of the Rice Terascale Cluster, a supercomputer
being built at Rice with funding from the National Science
Foundation (NSF) and Intel Corporation. This supercomputer
should be able to perform one trillion calculations per
second when it becomes operational next year.

For preliminary
results, Stoll and Subramanian are studying event data from
1979 to 2001 on eight countries in the Middle East. “We
know where and when the serious conflicts occurred, so we
can get a reality check on our predictions,” Stoll
said. Using a signal-processing technique called “wavelet
analysis,” they have discovered discontinuities known
as “singularities” in the events data that are
associated with the outbreak of serious conflict.

If the project
is successful, it could prove useful to policy-makers. Theoretically,
the information could be made available online, where officials
could consult it and possibly intervene to avoid conflict.
“Access to aggregated event data over a long period
of time can have a major impact on policy-making by providing
an additional source of information upon which to base foreign-policy
decisions,” Stoll said.

“But first
we have to address the core scientific question,” he
noted. “How well can an objective, data-driven approach
to modeling the genesis and evolution of conflict in various
regions of the world work?”

Subramanian and
Stoll acknowledge that a collaboration between a political
scientist and a computer scientist is rare, and they credit
the educational environment at Rice for making such a project
possible.

The database
of event information will be made available to the scientific
community.

The project is
funded by NSF and Intel.