Researchers have introduced an artificial intelligence system that can generate detailed, customized maps that may help governments predict how natural disasters will impact their communities.
The tool, detailed May 2 in the International Journal of Intelligent Systems, is especially useful for tracking pandemics, since it can account for population density, age demographics, commuter mobility and other factors that may contribute to the spread of infectious diseases, including the novel coronavirus. For the purposes of this study, De Montfort University researchers gathered demographic data from the United Kingdom mainland and applied it to the COVID-19 pandemic. But the team noted that users can easily eliminate or substitute these categories, depending on the nature of a natural disaster or a jurisdiction's access to a particular dataset.
The researchers hoped to address one of the most hotly debated aspects of government intervention during the pandemic — namely, to what extent leaders should shut down cities and states to prevent the spread of the virus. Even as early as the end of March 2020, over 100 countries had initiated lockdown measures that affected billions of people. But those lockdowns may have been more effective in quelling the spread of disease had they been tailored to the unique characteristics of affected regions, according to Raymond Moodley, a researcher at DMU's Institute of Artificial Intelligence and the paper's corresponding author.
"If you look at how the lockdown was implemented in the U.K., it was largely one-size-fits-all. The dynamics of London [are] not the same as, say, rural England, or coastal England, for that matter," Moodley told The Academic Times. "And if you look at the U.S. environment, for example, you can't apply New York City principles to Omaha, Nebraska."
AI-generated maps can help governments come up with highly customized guidelines for local districts, differentiating between jurisdictions that will be hit hard from a natural disaster and those that may go relatively unscathed. What's more, the regions least likely to be affected by a disaster can instead pool resources to support neighbors in areas that are harder hit.
The tool can also reveal unexpected geographic areas of interest that may not stand out as obvious hot spots for a pandemic. For instance, the model suggested that the medium-sized city of Leicester, where DMU is located, faced a higher chance of being severely impacted by the pandemic, despite its relatively smaller size and open layout compared with major metropolitan hubs. One factor that contributed to the increased threat was that the city has a higher percentage of intergenerational households. It means that young people are more likely to be in close contact with older adults — a population that has a higher chance of being severely affected by the virus.
And indeed, Leicester was hit particularly hard by the pandemic. Since the team developed their tool during the pandemic rather than before it began, they were not able to predict the spread of the disease before its emergence in the U.K. Yet, they had shown that their model could effectively reflect those real-world results, even with data that had been gathered before the pandemic became a significant issue in the country. "So we cannot claim that we told you so — because we didn't," Mario Gongora, a lecturer at De Montfort University and a co-author on the study, told The Academic Times. "But the nice surprise is that we realized that we could have told you so."
The researchers employed a concept called "self-organizing maps," grouping similar data into clusters without divorcing those statistics from their original context — in this case, the geographical locations they represent. "What we wanted to do was find similar areas within the U.K. that had similar characteristics, but at the same time, maintain that structure of the U.K. map, as opposed to extract them out of that environment," explained Moodley.
The model is not just for pandemics; it can be applied to a number of natural-disaster scenarios. Floods are one such interest for the team, who say they can plug floodplain data into the network to determine locations that will be most at risk. But there are some challenges that could arise when the AI is applied in international contexts, the researchers cautioned. Not every nation has the same access to the kinds of robust data that help power a neural network, and even the way information is inputted and logged can cause headaches for researchers as they attempt to integrate data into their system.
But for now, the team is aiming to utilize its system in the early stages of an upcoming natural disaster to see how accurately it can predict outcomes across regions. Their approach is ultimately an attempt to use theory to solve a practical issue. "The newness is taking an existing AI technique and applying it to a real life problem, and then coming up with high societal impact," Moodley said. "Being able to predict a pandemic, being able to predict a national disaster, is a great practical use of AI, and so that's essentially where we're driving [toward]."
The paper "Using self‐organising maps to predict and contain natural disasters and pandemics" published May 2 in the International Journal of Intelligent Systems, was authored by Raymond Moodley, Fabio Caraffini and Mario Gongora, De Montfort University; and Francisco Chiclana, De Montfort University and the University of Granada.