The structure and pace of the MQM program challenge us to stay consistent, work efficiently, and quickly absorb new conceptions. It is a rigorous program, and sometimes the tight timeline of our courses means we don’t have as much time to practice our new skills as we would like.

Even though it meant working on a big project during our most significant break from classes, we’re so glad we participated in the Fuqua Quantitative Management Programs Data Competition. The competition motivated us to find new applications for machine learning and data analysis in a field that we’re deeply passionate about: sports.

What motivated you to participate in the Fuqua Quantitative Management Programs Data Competition? How did it align with your academic and personal interests?

TONY: Advanced Analytics with Professor Jiaming Xu was an incredible course packed with cutting-edge machine learning concepts, covering topics like large language models (LLMs), deep learning, and more. We learned new tools and model structures in almost every session, making it a high-intensity learning experience. The class moved at a very fast pace, and I felt I didn’t have enough time to fully reflect on how to apply what we learned.

I left the course feeling slightly unfulfilled, knowing I had learned something incredibly valuable yet wishing I had more time to explore it in greater depth and apply it in a way that could have a real impact. The Fuqua Quantitative Management Programs Data Competition was the perfect opportunity to push myself to stay engaged with machine learning and continue researching its applications. It also allowed me to explore how machine learning could be applied to sports analytics.

The competition provided a structured challenge with a competitive element (which I found exciting!) and a well-defined timeline during winter break. It also encouraged us to consolidate our research into a practical application, create a demo, and document our progress.

As Robert and I planned our project, one particular case study from class stood out to me. I was interested in how machine learning is used in autonomous driving — specifically, how object detection algorithms track vehicle movements to avoid hazards. I realized that a similar approach could be applied to analyzing Duke basketball game footage. By systematically studying past games and leveraging object detection to recognize jersey colors, player shapes, and movement patterns, we could automate player tracking. This could not only save human labor but also provide real-time, in-game analytics that offer strategic advantages.

ROBERT: I always appreciate class materials that I can apply outside of class in a useful way, and that is how I felt for so many things I learned during my time in the MQM program. When the announcement of the Fuqua Quantitative Management Programs Data Competition came up, I knew that it was my time to prove my skills and bring my textbook knowledge to life.

Unlike Tony, I did not have a computer science background from undergrad, giving me an extra appreciation for the opportunity to learn about machine learning and other coding work along those lines. During the Fall term, I learned a lot from Professor Jiaming Xu and Professor Alex Belloni about machine learning using the Python and R programming languages. Professor Ryan Burk’s course also provided my classmates and me with knowledge of Custom SQL, which helped us get the most out of Tableau, which was learned later in Spring.

All these lessons were so refreshing and quite challenging to me. They laid the foundation for Tony and my work for the data competition.

What was the most rewarding aspect of this process? What skills did you develop or refine along the way?

ROBERT: Given the time that we had and the fact that we decided to create four new projects from scratch, time management and project management were extremely important in order to finish everything in time. We made a clear goal for each meeting and also relied on delegating different portions of our project. This way of working helped us to be efficient and keep each other accountable.

Another rewarding outcome of the Fuqua Quantitative Management Programs Data Competition was that it gave us the confidence to showcase our skills on a larger stage. During our spring break, we went to Boston for the MIT Sloan Sports Analytics Conference (SSAC). We participated in the hackathon, continuing our way of working together for a tight 6-hour timespan. Although we did not place in the end, we were praised by the judge and fellow competitors.

TONY: This process was an immersive learning experience that helped me develop new skills while building my confidence in what I was learning in the MQM program. Here are just my top three takeaways:

1. Effective planning and time management are crucial.

We only had one month to complete four projects. Robert and I met four times over Zoom during the break, and for each meeting, we clearly outlined our progress, goals, and overall timeline. I’m incredibly proud that we stuck to our initial schedule throughout the entire process, despite traveling extensively during the break.

After advancing to the final round, the new semester had already begun, so we had to balance our academic coursework with preparing for our final presentation. During the last week before the presentation, we often stayed on campus until 1 a.m., dedicating significant time to refining our Duke Basketball Player Detection project and rehearsing extensively. With only eight minutes to present all four projects, we had to ensure seamless transitions between sections and keep the audience engaged without overwhelming them with information, putting emphasis on time management in an entirely different way.

2. Data preparation is just as important as the model itself.

Through this experience, I gained a much deeper understanding of applying advanced machine learning tools in real-world scenarios. One key lesson I learned was the importance of data collection — something that’s often overlooked in academic projects. In school, datasets are typically pre-cleaned and formatted by professors, making them ready for immediate use. However, in our projects, we quickly realized how challenging it is to find a clean, recent, well-documented, and comprehensive dataset from free, public sources.

Additionally, implementing machine learning algorithms is often not a hard part of the process. The challenge lies in organizing and preparing the data. Cleaning the dataset, adjusting formats, creating dummy variables, and converting between string and numeric values all require experience and significant time investment.

3. Overcoming challenges requires resilience and thinking outside the box.

For our Duke Basketball Player Detection project, we initially trained our model using free basketball datasets from Roboflow. We tested at least five of the most widely used datasets for sports player recognition, but the results were far worse than expected. At one point, I felt frustrated and directionless.

After taking a step back to reflect, I realized that since we were working with Duke basketball footage, we needed a customized dataset to achieve our specific goal. I spent two days learning how to create our own sports vision dataset on the Roboflow platform and retrained our models using our newly curated training set. The performance exceeded our expectations, and the model started correctly identifying Duke players and their opponents.

This experience reinforced the critical role of data collection and taught me the importance of thinking outside the box. Instead of settling for pre-existing datasets, I learned that sometimes the best approach is to build a tailored dataset to fit the project’s unique needs. It was a valuable lesson in problem-solving and adaptability.

How has your perspective on data analytics changed since joining Fuqua?

ROBERT: Before coming to Fuqua, I mainly used simple tools for data analysis and visualization, like Excel. I knew that data was important, but making it useful or gaining deep insights on a high level was a challenge.

As I near the end of my MQM experience, it’s clear that I have learned a lot of tools that I can utilize to make data more understandable by the right audiences. Besides the analytics part, there are also storytelling, data cleaning, and so many other important things that I find extremely useful.

TONY: One of the most valuable lessons I’ve learned is the power of simplicity in data analytics — how using simple models and creativity can provide deep insights into even the most complex problems. This realization has completely transformed the way I approach data analysis.

At one point in the program, I was fascinated by fine-tuning parameters and applying advanced machine learning algorithms to every project. The excitement of improving a model’s accuracy from 7.4% to 7.3% gave me a strong sense of achievement. However, I became overly focused on squeezing out incremental numerical improvements while neglecting a crucial question: Is this truly valuable or interpretable in a real-world context?

Participating in the winter competition and taking the Marketing Intelligence course with Professor Allison Chaney shifted my perspective. I started to appreciate the power of simplicity. Instead of prioritizing complex models and cutting-edge algorithms, I realized I should first focus on understanding the core problem and identifying the best way to interpret or address it.

To me, data analytics is ultimately a logical thinking process. The numbers, findings, and models should align to support a clear, compelling idea. While I have access to a wide range of analytical tools, the real skill lies in choosing the right tool based on the message I want to convey.

Now, I spend more time on reflection, communication, and research, ensuring that my analysis is meaningful, rather than obsessing over model performance alone.