One of the most common types of datasets used is health/medical data, as it allows students to analyze topics that can have real-world impact. For example, one group of students obtained de-identified medical claim records from a large insurance provider covering several years. They analyzed the data to identify predictors of high medical costs and develop risk profiles that could help the insurance company better manage patient care. Some features they examined included diagnoses, procedures, prescriptions, demographics, and lifestyle factors. They built machine learning models to predict which patients were most at risk of future high costs based on their histories.

Another popular source of data is urban/transportation planning datasets. One project looked at public transit ridership patterns in a major city using anonymized tap-in/tap-out records from the city’s subway and bus systems. Students analyzed rider origins and destinations to identify the most traveled routes and times of day. They also examined how ridership changed on different days of the week and during major events. Their findings helped the city transportation authority understand demand and make recommendations on where to focus service improvements.

Education data is another rich area for capstone work. A group worked with a large statewide standardized test scores database containing student performance dating back over 10 years. They performed longitudinal analysis to determine what factors most strongly correlated with improvements or declines in test scores over time. Features they considered included school characteristics, class sizes, teacher experience levels, as well as student demographics. Their statistical models provided insight into what policies had the biggest impacts on student outcomes.

Some students obtain datasets directly from private companies or non-profits. For example, a retail company provided anonymous customer transactions records from their loyalty program. Students analyzed purchasing patterns and developed segments of customer groups with similar behaviors. They also built predictive models to identify good prospects for targeted marketing campaigns. Another project partnered with a medical research non-profit. Students analyzed their database of published clinical trials to determine what therapies were most promising based on completed studies. They also examined factors correlated with trials receiving funding or being terminated early. Their analyses could help guide the non-profit’s future research investment strategies.

While restricted real-world datasets aren’t always possible to work with, many students supplement private data projects with publicly available benchmark datasets. For example, the Iris flowers dataset, Wine quality dataset and Breast cancer dataset from the UCI Machine Learning Repository have all been used in student capstones. Projects analyze these and apply modern techniques like deep learning or make comparisons to historical analyses. Students then discuss potential applications and limitations if the models were used on similar real problem domains.

Some larger capstone projects involve collecting original datasets. For instance, education students designed questionnaires and conducted surveys of K-12 teachers and administrators in their state. They gathered input on professional development needs and challenges in teaching certain subjects. After analyzing the survey results, students presented strategic recommendations to the state department of education. In another example, engineering students gathered sensor readings from their own Internet-of-Things devices deployed on a university campus, collecting data on factors like noise levels, foot traffic and weather over several months. They used this to develop predictive maintenance models for campus facilities.

Real-world datasets enable capstone students to gain experience analyzing significant problems and generating potentially impactful insights, while also meeting the goals of demonstrating technical and analytical skills. The ability to link those findings back to an applied context or decision making scenario adds relevancy and value for the organizations involved. While privacy and consent challenges exist, appropriate partnerships and data access have allowed many successful student projects.