---

When Esha Singaraju first arrived on campus in 2023, she wasn’t sure what path she wanted to follow. Like many first-year students, she explored a range of interests, enrolled in various classes, and searched for her niche.

One class unexpectedly shaped her academic journey: “Environmental Law and Policy,” taught by Carolina law professor Donald Hornstein.

Through this course, she learned about climate legislation and the Federal Emergency Management Agency (FEMA), which supports citizens and emergency personnel in preparing for, responding to, and recovering from a wide range of hazards.

“Natural disasters are so volatile,” Singaraju says. “You don’t know how much they are going to affect you, but to be able to use data from past disasters to help prepare for future ones is a really cool idea.”

Now a junior at Carolina, she is collaborating with UNC-Chapel Hill professor Gregory Characklis and the UNC Institute for Risk Management and Insurance Innovation. This multidisciplinary hub brings together researchers and industry professionals to develop solutions for managing the financial risks faced by communities worldwide, ranging from extreme weather events to cybersecurity threats.

Singaraju is working on a project to collect financial aid data from FEMA and the U.S. Small Business Administration (SBA), a federal agency that provides financial assistance and counseling to help Americans start, build, and grow businesses. Her goal is to develop machine learning models that predict the amount of aid North Carolina homeowners may receive after future natural disasters.

“What excites me about this project is the blending of technology like AI with important topics like climate policy,” she says. “It’s interesting to see how powerful data can be in addressing a real-world problem.”

Seeking aid

In North Carolina, approximately 97% of homeowners lack flood insurance, according to the state’s insurance commissioner. This becomes a critical problem after disasters like hurricanes or floods, when homeowners are left searching for financial solutions.

“In the past 20 years, you’ve seen all these hurricanes and flooding, and that has had a big impact on homes in the state,” Singaraju says.  “And then it opens up to some disparities in how this federal aid is distributed.”

There are two federal aid sources designed to help homeowners recover after a disaster has been declared by the federal government. Run through FEMA, the Individuals and Households Program (IHP) provides grants to homeowners to help them pay for temporary housing, repair or replace their damaged homes, and cover other disaster-caused expenses.

The SBA leads a second program that offers low-interest, long-term disaster loans to homeowners to help with repairs or replacement of damaged homes.

“These programs are essential for North Carolina homeowners, but the application and approval process can be complex and vague,” Singaraju says.

The common reasons cited for IHP denials include insufficient damage — when the applicant’s losses did not meet FEMA’s threshold for eligibility — incomplete or missing forms, and other documentation errors. But it is not always clear why an application is denied or approved.

SBA disaster loan approval or denial, as well as the loan amounts, can vary and may not be fully understood by homeowners.

“There’s so much uncertainty in how or why someone gets approved and how much they’re even likely to get approved for,” Singaraju explains.

She saw an opportunity: What if she could use artificial intelligence to replicate the decision-making processes used by these agencies? And if so, could she predict which homeowners would be likely to be approved for disaster aid?  Could she identify key factors that affect these approvals?

The answers could help clarify and streamline this confusing process.

Building the algorithm

During the summer after her freshman year, Singaraju joined Characklis’s lab and began developing her algorithm — an incredibly ambitious goal considering she’d only taken one computer science course.

“I had no idea how to do any of it,” she recalls with a laugh. “It was really a 0 to 100 process of doing research, but also not knowing how to apply these tools just yet. It was a lot of learning how to code, how to think about designing research, and asking good questions.”

Along with learning the basics of coding, she also needed to find the data. For the IHP, FEMA provides downloadable website data that includes details from applications such as approval status, aid amounts received, and the extent of home damage. In contrast, the SBA is more restrictive with its data, making it challenging to obtain.

To overcome this barrier, Singaraju collaborated with a master’s student at the UNC Institute for Risk Management and Insurance Innovation (IRMII) to submit formal data requests to the SBA. During their search, they came across a research paper that used the association’s data and reached out to its authors, who generously shared their dataset.

“IRMII is so incredibly hands-on because everyone in that organization is so passionate and interested in their research,” she says.

She was invited to participate in an informal flooding research group composed of graduate students and faculty. During these biweekly meetings, she received feedback on everything from research methods to coding.

Singaraju spent most of last summer and fall working on the predictor model for the IHP program. She’s currently working on refining the model for the SBA program and hopes to combine the two in the coming months.

While the project is a work in progress, she hopes that it will be a useful tool for homeowners, researchers, and policymakers.

“I want to really transition this from something that’s research-centric into something that’s impactful and useful to people,” Singaraju says. “I’m someone who likes to think ahead and then think logically and critically. And to be able to do that on a computer using code and then seeing tangible results of how it’s impacted somebody is exciting.”