How Portland State University accelerates student degree completion through machine learning
Portland State University (PSU) is using machine learning (ML) to help students find the most effective pathways to graduation. By tracking the course history of successful graduates and presenting recommendations to current students, PSU can provide guardrails and best practices for a focused journey towards degree completion.
With the emergence of massive open online courses (MOOCs), Nano degrees, and online programs, innovative higher education institutions such as PSU are evolving to meet the changing demands of students and deliver more value to compete with peer institutions.
A priority for PSU is to become a data-driven institution to facilitate development across the university and improve student outcomes. PSU is working to help students reach their educational goals faster. According to the National Center for Education Statistics, only 60 percent of college students receive a degree within six years. PSU is looking to change that for its students by using ML as part of their efforts to improve student success.
To start, the PSU team tackled the challenge of making sure students registered for classes that aligned to their degree goal. Students often have difficulty navigating course offerings to decide which courses will lead to successful completion of their declared major. Taking additional courses outside of a student’s declared path can be disruptive to the graduation timeline and lead to attrition. To help deliver value and facilitate student success, PSU wanted to provide students with proven pathways to complete their degree requirements while also enabling the university to plan for its future, including course and staffing requirements. Secondly, this initiative served to strengthen the university’s competency with data science to help address additional projects over time.
The PSU and AWS Professional Services (AWS ProServe) team developed ML models to enable the following:
Using transfer students’ course histories to predict the best alignment for degree plans. The PSU student body has a high percentage of students who transfer (50 percent) from other schools. The team developed an ML model from historical student data with the intent of informing transfer students of the quickest path forward and suggested majors based on their prior course work. This model simplifies the task of searching through the catalogue to find the right match of courses for each major’s requirements.
Alerting students when they choose courses that deviate from their degree plan. This model calculates whether the course selection deviates from the usual range of courses taken by students who successfully completed that degree.
Assisting in capacity planning based on the paths of students currently enrolled. The predictor model is applied to a cohort of students in any particular year to predict the number of students (both declared and non-declared majors) in future years to better estimate the number of students in each major to help with resource planning.
As with most ML projects, the team worked through steps to understand and prepare the data for training the models. Due to the volume of inbound student transfers, there was diversity in the paths leading towards a degree and models based on diverse datasets lose predictive power. However, by identifying majors that had meaningful correlations across introductory courses, the algorithm was able to suggest likely majors with reasonable accuracy. Another issue arose as the models identified certain majors where few students completed the degree plan, which resulted in models with an innate bias against these majors. Future work and modeling will be required to address this issue.
The team discovered a sufficient number of these unusual cases and recommended that a human advisor be in the loop to guide student decisions. Although there are long-term aspirations to share the recommendations with students directly, the most effective application currently will be through academic advisors.
“PSU’s commitment to putting students first demands that we employ the best information to ensure that we are using our resources most effectively in supporting students to achieve their educational goals. We’re really excited to test how ML can be an active component of that information landscape,” said Susan Jeffords, provost and vice president for academic affairs, PSU.
The PSU team plans to build on this project to implement improvements such as incorporating student success data and potentially incorporating information on graduates. They plan to make alerts and reports for students available to academic advisors to increase their effectiveness in helping students make strategic decisions. The guided pathways model will give advisors a new source of information highlighting when students select courses that are outside their best track towards their declared major and may then contact the students to influence a successful outcome.
An important consideration discussed with PSU was whether to include student success within their choices for majors. Available data included grades in classes, which could be included to improve suggestions of courses, such as the best order to take certain required classes. In addition, this data could help suggest courses that may not be in the major requirements but could assist certain students to be successful in future courses. This type of information is available to students at a high-level, but a prediction ML model could automate the process of using this data to help students make informed decisions.
To learn more about PSU’s story, listen to the AWS Fix This podcast episode, Keeping students on the right track. To engage with AWS ProServe, visit the AWS Professional Services webpage, and view the machine learning service page to see other examples of how customers are using this unique technology.