Using AI to redesign the college degree attainment experience to improve graduation rates for disadvantaged students
In Winter 2019 and at Stanford’s Designing AI to Cultivate Human Well-Being class, interdisciplinary teams of students worked together to explore solutions to several important societal problems through the application of technology and AI. In this post and in the few following ones, the teaching team will be highlighting the top 3 teams and how they defined their problem statement as well as their projects’ outcome.
Overall, just 57% of college students obtain a degree after six years. At four-year for-profit colleges, the typical completion rate is 35%. The situation is a little better at public (two-year) community colleges, where the completion rate is just 38%.
So, How can we increase completion rates?
Here is how the team approached the problem in their own words:
We began our research by examining the data on the problem. Data about dropout rates is available in final form (cleaned, organized, and presented) from the federal government’s National Center for Education Statistics,3 think tanks, and academics. But our interview with Stanford Professor Ari Kelman revealed that these statistics actually masque the complexity of what is actually being asked. To clean this data, sources need to define a drop-out. We had in our mind the student who, from stress or lack of social connection, slowly fades out of education. But the reality is that there are many productive reasons why a student might choose to drop out of higher education. But on the other end of the spectrum are the Bill Gates and Mark Zuckerberg, who, while outliers, also count as dropouts. More complex, and much more likely, are the students who fall somewhere in the middle — they might leave to take care of an ailing members of their family, or to pursue different paths to a career; we ascribe a positive normative value to their intentions but the economic effects of dropping out persist.
Our proposed solution focuses on using AI to both better identify at-risk students and optimize intervention strategies. Students provide a plethora of personal information about themselves: in their college applications, social media activity, and collegiate activities. Our hypothesis is that these sources could be consolidated to better understand how student college experiences differ and which students are actually at risk of dropping out. Our solution would take this myriad of factors and their relative weight in affecting the probability of a student dropping out and assign a gradient risk score for each student. Based on how different factors influence a specific student’s “risk score,” the school’s intervention can change to better meet the student’s personal needs at that time. This kind of analysis can be accomplished using a regression and classification model in order to categorize students into at risk, and safe groupings.
During the last class session, the team delivered their final presentation with their key insights and takeaways. Let us take a look:
References: College Completion Rates Are Still Disappointing