Using Machine Learning To Predict Food Insecurity

The lives of hundreds of millions of individuals across the globe are at risk due to food insecurity. The Food and Agriculture Organization of the United Nations reports that the number of undernourished individuals has risen from 624 million people in 2014 to 688 million in 2019.

A new study from New York University explains a new machine learning model that can accurately predict regions that are at risk of food insecurity by analyzing the content of news articles. This model offers an improvement over current measurement methods and could assist in prioritizing the distribution of emergency food aid to vulnerable areas.

“Our approach could drastically improve the prediction of food crisis outbreaks up to 12 months ahead of time using both real-time news streams and a predictive model that is simple to interpret,” the researchers explain.

Food insecurity

“Traditional measurements of food insecurity risk factors, such as conflict severity indices or changes in food prices, are often incomplete, delayed, or outdated,” they continue. “Our approach takes advantage of the fact that risk factors triggering a food crisis are mentioned in the news prior to being observable with traditional measurements.”

Although food insecurity is a prevalent and pressing issue, current methods of detecting future food crises are inadequate and limit the ability to address them effectively.

To create a better predictive model, the researchers explored the possibility of utilizing news coverage, which provides real-time accounts of local events, as an early warning system for impending food crises.

To this end, the researchers analyzed over 11 million news articles covering approximately 40 food-insecure nations published between 1980 and 2020. They devised a technique to extract specific phrases from the articles related to food insecurity, taking into account the journalistic evaluation with notable precision.

Detecting crises

The current methods to detect future food crises are inadequate, despite the widespread and urgent nature of the problem. Researchers aimed to develop a better model by using news coverage as an early-warning system for impending food crises. To achieve this, they collected text from more than 11 million news articles focused on almost 40 food-insecure countries published between 1980 and 2020, and extracted particular phrases related to food insecurity using a tool that accounted for nearly 170 text features.

The researchers then examined data on various food-insecurity risk factors to determine if there was a correlation between news mentions of these factors and their occurrence in the studied countries and regions. They found that news stories were an accurate indicator of the studied conditions, as there was a high correlation between the nature of the coverage and the on-the-ground occurrences of these factors.

To determine if news articles were a good predictor of subsequent food crises, the team compared the predictive accuracy of news coverage against traditional measurements. They found that news coverage yielded more accurate predictions at the local level of food insecurity than traditional measurements that did not include news story text, up to 12 months ahead of time, across 21 food-insecure countries from 2009 to 2020. Supplementing traditional predictive measures with news coverage further improved the accuracy of food-crisis predictions, suggesting the value of “hybrid” models.

The researchers see potential for their work to be used on a larger scale to help prioritize the allocation of emergency food assistance across vulnerable regions.

“News indicators could be extended to the prediction of disease outbreaks and the future impact of climate change,” they conclude.

Facebooktwitterredditpinterestlinkedinmail