Using AI to Identify At-Risk Students
Most colleges and universities have a lot of student data available in their systems, but they may not have the resources to easily analyze that data for meaningful use. Florida International University contracted with DataRobot to leverage its machine learning suite to test “every possible predictive methodology,” says Hiselgis Perez, FIU’s associate vice president of analysis and information management. “Utilizing the DataRobot platform turned what was an extremely long and complex task into something simple and straightforward.”
The FIU models consider each student’s academic performance, programs, campus engagement and other variables to give each student a probability score for retention, graduation and other outcomes, Perez says. Any student with a probability score below 70% is considered at risk.
“The hard work is then taking those insights, such as lists of students or courses, and working with the responsible parties to address those issues,” Perez says.
For example, if the model references a student as being at risk due to low GPA, that student may be referred to advising or academic support programs. The model may also highlight roadblocks in the curriculum when examining student flow through courses, which could necessitate course-level interventions such as optimizing instruction or adding more sections.
DISCOVER: Four AI trends to watch this year.
Combining AI Tools and Human Intervention
Similarly, Georgia State University uses predictive analytics tools to track more than 800 risk factors for every student daily. But tracking the data is only the beginning; combining the findings with human intervention is what makes the difference for student outcomes, Renick says.
For example, when a student drops a class in the middle of a semester, most universities do not systematically reach out to that student. But Georgia State uses technology to not only flag that at-risk behavior but also notify an adviser to reach out to the student.
“When we act on that behavior in the moment, we often can help the student deal with the larger problem that may be happening,” Renick says. “Maybe they dropped the class because they’re going through a breakup, experiencing financial strain, or they’re just overwhelmed. We can help them on the spot to address the underlying issue.”
Academic advising has traditionally been a “rather passive enterprise, relying on students to self-identify their problems and take it upon themselves to ask for help,” Renick says. However, because many students never diagnose their own problems or navigate campus bureaucracy to get the help they need, advising may not happen before it’s too late. By combining predictive analytics with proactive intervention, Georgia State has improved graduation rates by 7% among the total student population, with even stronger gains for students from underserved backgrounds, Renick says.
UP NEXT: A unified data ecosystem can support AI initiatives.
Leveraging Chatbots for Personalized Feedback
With so much student data available, it can be challenging for institution staffers to reach out to every student who may exhibit at-risk behaviors. That’s why AI-based chatbots can be helpful. Georgia State, for example, built a knowledge base of answers to 500 commonly asked questions, along with an algorithm that can locate the answer or send the question to a staff member if needed.
“Our chatbot has helped improve retention, helped students complete the Free Application for Federal Student Aid (FAFSA) on time, and helped students meet with advisers more regularly,” Renick says. “In a focus group, students said they asked the chatbot questions they wouldn’t have asked a human being.”
At FIU, on-demand chatbots support students by asking a number of different questions and integrate with Canvas courses so students can ask about course content, Perez says. For example, a student could go to the chatbot to get more clarity on the syllabus or on the expectations for a particular assignment.
“Most institutions have the data available to them to do similar modeling to what we have done,” Perez says. “I would urge them to look closely at the information that they have on hand and think about how those data points could tell the story of your students. A significant portion of the science of machine learning is also the art of telling the story of your students and your institution through data points and analyses. The best approach is always going to be one that serves the students in the best way possible.”
