The Boston Public Health Commission’s Center for Public Health Science and Innovation (PHSI) works across city programs to apply advanced analytics and innovative data methods to inform public health decision making. By combining operational data with new analytics tools, the PHSI team helps public health leaders anticipate service needs, allocate resources effectively, and better support Boston residents. But for some public health services, demand can vary significantly from day to day, making it challenging to predict the right levels of staffing, space, and supplies needed across the Commission.
Predicting the unpredictable
“Many public health services operate in environments where demand can change quickly,” says Dr. Sanouri Ursprung, director of the Boston Public Health Commission’s Center for Public Health Science and Innovation. “Using novel methods, and a wider range of indicators to anticipate those changes can help programs prepare more effectively and ensure services remain accessible for the Boston residents who most heavily rely on them.”
For example, BPHC’s shelters are easily accessible, and open 24 hours a day, seven days a week. They don’t turn away anyone who needs assistance and demand can fluctuate from night to night. Anticipating potential increases can help staff better plan for space, staffing, and other resources.
Using this example, BPHC engaged with a team of students taking MIT IDE’s Analytics Lab (A-Lab) to explore how a wide range of non-traditional data such as weather metrics, unemployment statistics, and rental patterns can create a predictive model to forecast demand for public health services.
“Traditional analytic approaches often rely on a limited number of datasets because integrating and testing so many indicators can be prohibitively time consuming when done manually. Also, many disproportionately impacted populations are underrepresented in traditional health data systems, which can make it harder for public health agencies to anticipate and respond to their needs,” says Dr. Ursprung. “Using AI and machine learning to explore a broader set of indicators can help us quickly refine prediction models for public health services and better tailor support for populations that are often invisible in conventional data sources.”
The A-Lab team was enthusiastic about the project.
I was drawn to this project because I was excited about the idea of working with a government agency to promote societal good. In the business analytics space, the end goal often boils down to increasing profits, and it was refreshing to be able to contribute to the public sector.
Master of Business Analytics student Jérémie Taranto, MBAn ’26, was inspired by his host’s enthusiasm, and the positive impact the project could make on the Boston community.
Seeing how data could directly support public health decisions made me want to be part of it.
Balancing accuracy and flexibility
Excited to apply their knowledge to this challenge, the students set to work creating a tool that would best help BPHC predict demand for public health services. They tested different combinations of models that allowed BPHC to input data along with a wide range of publicly available indicators to compare results. The team’s goal throughout testing was to maximize accuracy while also maintaining flexibility in the face of widely fluctuating conditions.
The A-Lab team found that the best approach was to create a dynamic model selection framework. This framework compares predictions from the two best performing models—an AR-based census model and an Ensemble model—and chooses the one with the least errors. The pilot model then makes daily demand predictions for the next fourteen days.
When we tested different models, we initially expected complex methods to outperform simpler approaches. However, we realized that using a minimal number of optimally engineered features could generate better, more interpretable results.
In the end, the students created a prototype tool that allows BPHC to easily input the data of their choosing and visualize the model’s predictions. Rather than relying on a fixed set of variables, the team designed a dynamic modeling framework that allows BPHC to adjust the data inputs over time. This flexibility enables the public health team to incorporate new indicators or contextual information as conditions evolve, without rebuilding the model from scratch. Furthermore, the pilot tool does not store any underlying data, allowing Commission analysts the flexibility to adjust inputs and integrate new data sources as needed in the future.
Ursprung, whose team will soon begin testing and refining the tool, believes this project work has the potential to inspire other jurisdictions by demonstrating the power of connecting Boston’s innovation ecosystem with public sector leadership.
“Boston is unique in that it is home to an immense wealth of academic and private sector innovators,” she says, “It is also home to public health civil servants who are nationally recognized for their commitment to health equity. By connecting these two groups, we showcase how new analytic approaches can be applied to real world challenges to ensure they improve the health and well-being of all Boston residents regardless of age, race, gender, housing status, etc.”
She hopes the project will help demonstrate how engagements between universities and public sector organizations can apply new data science approaches in ways that prioritize the public good and already has plans to share the lessons learned from this pilot with local and national public health partners.
“Too often these kinds of analytic tools are developed primarily for commercial applications,” she says. “By bringing together academic expertise and public health practitioners, we are showcasing how these partnerships and innovations can be used to support communities that are often invisible or underserved in traditional data systems.”
Making a positive impact with data
Inspired by their project, the A-Lab students are motivated to continue using data analytics to help others.
“Having a positive impact on an issue like shelter services has helped me reflect on how I want to use my degree in the long term. It was inspiring to see the passion and care that the BPHC team brings to their work and has opened my eyes to a lot of good that could be done by doing data analytics in organizations and fields that promote the public good,” says Hoch.
Data work in the real world doesn’t start with models—it starts with understanding the problem and the people involved. The technical piece matters, but it’s only one part of a much larger process.
