Data Analytics in Action
From saving lives to predicting prices, 2024 MIT A-Lab winners solved tough business problems with advanced data analytics.
By Peter Krass
Accurately diagnosing a medical condition can be a major challenge for physicians, but treating the disease can be tricky, as well. How does a physician determine which medication will be most effective for the patient, especially when there can be more than a dozen appropriate medications to choose from?
The answer lies with data analysis, says a team of MIT graduate students. Calling themselves Vitalytics, the team — Nidhish Nerur, Hunter Sporn, Jaeyoon Wang and Naiqi Zhang (pictured above) —combined clinical and biomedical knowledge graphs provided by their corporate partner, Sanofi.
Their goal: help physicians find other patients with similar symptoms who responded well to a specific drug. “Together,” said team member Sporn, “we can improve patients’ lives.”
The Vitalytics team’s project, “Patient Like Me” mentored by Tomasz Grzegorczyk, was impressive enough to win this year’s MIT Analytics Lab. “This team and their project were so strong,” said Renée Richardson Gosline, a Senior Lecturer at the MIT Sloan School and one of the award’s three judges. “It really stood out on the merits of the technical exposition, the effort made by the team, and the solid presentation. Plus, the impact was undeniable.” The team will have their names engraved on a large silver trophy.
Rising to the Challenge
This year’s Analytics Lab presentations on December 13 culminated the fall semester’s action learning program. Better known as A-Lab, the course is spearheaded by the MIT Initiative on the Digital Economy (IDE) and run by the MIT Sloan School to connect students with business challenges.
During its 11 years, A-Lab has attracted some 800 students from a dozen MIT departments. They’ve worked on projects spanning IoT, digital platforms, finance, marketing, e-commerce, retail, manufacturing, medical supply chains, workplace safety, and global health. Project sponsors this year included Wayfair, Duolingo and MassMutual. The companies proposed the challenge and provided the 20 teams with corporate data for analysis and potential solutions.
Each A-Lab team, typically composed of four students, applies data analytics, machine learning or other digital experiments to solve the problem. They’re helped by mentors, many of whom are former A-Lab students in the friendly competition.
In addition to the Sanofi project, teams presented solutions ranging from foreign-language learning, ways to improve retail promotions, and even training to help war-zone civilians stay safe around explosive landmines.
The projects were judged on four criteria: technical and analytic; effort and contribution; business impact; and the quality of the final presentation. Three judges evaluated the work: Tod Loofbourrow, CEO of ViralGains; Michael Schrage, a Visiting Fellow at the Imperial College Business School, and Gosline of MIT.
Tomorrow’s Prices Today
The second-place winners, called the AnalytiXAvengers, worked with Gordian, a provider of data-driven services for all phases of the building life cycle.
Their challenge: evaluate and develop new methods for predicting the prices of various construction materials in volatile markets.
Gordian already uses a forecasting model that covers 70 materials and nearly 750 locations; it includes data going back 25 years. But the company wondered if different approaches could improve the model’s performance during turbulent times such as the COVID pandemic.
It could. The students combined machine learning technology, a rich dataset and domain knowledge to produce forecasts that were, on average, better by 8.5% during highly volatile periods. “Seasonality does exist,” said team member Badiss Ben-Abdallah. “But you have to look under the hood” to wring more performance from prediction models.
How Loyal is Loyal?
Third place went to team DeepInsights that worked with Cognira, a company that offers sales-promotion tools for retailers. The students were asked to answer a challenging question: Will loyalty-card holders respond more favorably to a retailer’s sales promotions than consumers who don’t hold the loyalty cards?
Using an MIT-developed technology known as “double machine learning,” they created a model that could predict the causal effect of a promotion for their project “Causal Estimation of Retail Promotions Effect on Sales.” Employing causal inference, they found that
on average, loyalty cardholders respond to retailers’ promotions seven times more than consumers without the cards.
However, that wasn’t true for all products. In one case, the promotion led to cardholders buying 64% more than non-cardholders while in another example, the results were negative, meaning non-cardholders actually bought more.
On presenting the bronze award, Schrage described the team’s in-depth work as a “good analysis that was very well done.” That’s a great description of the entire A-Lab course.
Peter Krass is a contributing writer and editor to the MIT IDE.
