Research may help planners predict food shortages

Posted by
On June 30, 2008

As world leaders seek ways to address global food shortages, Missouri
S&T researchers are working on a method that could help government planners
and relief agencies better prepare for future shortages by predicting the
variability of food supplies for specific nations or regions.

Employing a statistical method used by manufacturers to improve product
quality, Missouri S&T graduate student Parthiv Shah and his advisor, Dr.
Elizabeth Cudney, evaluated two years’ worth of agricultural yields from a
dozen industrialized nations to predict the output for a third year. The
results of their research were accurate to within 95 percent of the actual
yields for the third year.

Food shortages have been the subject of recent discussions of global
planners, as conferences in Rome and Paris in early June called for more
research on long-term agricultural sustainability.

“Using just two years of data, we are able to get fairly accurate
predictions with this method,” says Cudney, an assistant professor of engineering management and systems engineering
at Missouri S&T.

The method in question is known as the Mahalanobis-Taguchi System, one of
several statistical techniques devised by Genichi
, a Japanese engineer and statistician. Taguchi’s methods were
developed 50 years ago to help manufacturers test multiple aspects of a process
using just a subset of the tests that had been traditionally required. MTS is
also named after Prasanta Chandra Mahalanobis, who introduced a method for
measuring correlations between variables and the different patterns that can be
identified via this approach.

The system Shah and Cudney employed has also been used by manufacturers to
improve product design and customer satisfaction. Other Taguchi techniques have
been used by companies such as to predict the performance of
various websites, as reported in Ian Ayres’ 2007 book “Super Crunchers.”

But the method is considered controversial by some statisticians, Cudney
says, because it relies on small data sets to forecast results.

Controversial or not, the system worked well for Shah and Cudney’s
experiment. Their estimates of food supplies for 12 nations were extremely
accurate, despite being based on only two years’ worth of data from just 12
categories of agricultural products.

Working with 2001 and 2002 data on the yield of 12 types of agricultural
products – including grains, wheat, meat and dairy products, and fruits and
vegetables – the researchers were able to forecast 2003 agricultural yields for
a dozen nations with 95 percent accuracy.

According to Cudney, this same method could be used to help the World Bank,
government planners and relief agencies better forecast where food shortages
may occur in future years.

Food shortages have been the subject of recent discussions of global
planners, as conferences in Rome and Paris in early June called for more
research on long-term agricultural sustainability. The meetings followed
a May 29, 2008, report
from the U.N. Food and Agriculture Organization and
the Organisation for Economic Co-operation and Development, based in Paris,
that suggested policymakers reconsider biotech or genetically modified crops to
improve crop yields.

As it now stands, the Missouri S&T researchers’ model is relatively
simple. It uses only a dozen factors – the yields of a dozen different food
products – and pulls that data from a dozen industrialized nations, whereas the
U.N. and World Bank concerns lie more with food shortages in developing
nations. Moreover, the technique doesn’t take into account the impact of
imports, exports or weather patterns.

But the method could be expanded to incorporate those factors, the
researchers say. More complex issues related to supply chain – the movement of
goods from farm to market and across borders – could also be factored in to a
system to create a more sophisticated forecasting model, they say.

To test MTS beyond product development, Shah and Cudney conducted an earlier
study to forecast rainfall based on earlier data. Working with engineering
management student Vivek Jikar; Dr. David Drain, assistant professor of
mathematics and statistics at Missouri S&T; and Hiroshi Shibano of Japanese
printer manufacturer Konica Minolta, the researchers used monthly precipitation
data for a particular city from 1970 to 2006 to forecast future rain amounts.
Analyzing data from 1970 through 1999, the researchers were able to accurately
forecast rain amounts with 87.7 percent accuracy.

In both studies, researchers used a new technique, known as the “T-Method,”
which predicts outcomes by determining a signal-to-noise ratio for a process.
Signal-to-noise ratio is a term borrowed from electrical engineering, where it
refers to the level of a desired signal, such as music, to background noise.
With the Taguchi Method, the term refers to important vs. less important
information to determine outcomes.

The T-Method is virtually unknown and unpracticed in the United States,
Cudney says. “There was no literature about the T-Method in English,” she says.
“It all had to be translated from the Japanese.” Dr. Shih-Chung Tsai of General
Motors’ Technical Center in Warren, Mich., assisted Shah and Cudney by
translating the materials.

As a Ph.D. student at Missouri S&T, Cudney employed MTS as part of a
research project to help General Motors relate a vehicle’s technical
performance to improve customer satisfaction.

Share this page

Posted by

On June 30, 2008. Posted in Research