Chesterfield County Public Schools uses machine learning to predict county’s chronic absenteeism

Chesterfield County Public Schools (CCPS) in Virginia uses machine learning (ML) on Amazon Web Services (AWS) to predict the county’s rate of chronic absenteeism in high schools. CCPS includes 64 schools and over 63,000 students.

Morgan DeBusk-Lane, PhD, is the research and evaluation coordinator for CCPS. In his role, Dr. DeBusk-Lane oversees three main areas of data analytics for CCPS (research, evaluation, and assessment), all with the goal of providing predictive analytics to school practitioners, leadership, and the school board. “It’s not common in K12 to have predictive analytics, even though there is a practical need for it. Migrating our analytics off of our servers and onto the AWS Cloud helped us to move faster and be more agile,” says Dr. DeBusk-Lane.

Improving the K12 learning environment

Dr. DeBusk-Lane wanted the school leadership to begin to see the value in performing predictive analytics, which could help the K12 district improve the learning environment for students. He decided to start with a high-impact project that would use data the school system was already capturing about its students. Working with division leaders and other content specialists, they decided to tackle predicting chronic absenteeism first.

Chronic absenteeism can have long-term, negative effects on students’ economic and social development, and students who qualify miss between 10-100 percent of a calendar year’s school days. “We chose this use case first because it affects every school district, including ours. Chronic absenteeism leads to poor outcomes for students—failing classes, lower test scores, and higher drop-out rates. This model is meant to provide a way for schools to intervene early and to appropriately optimize intervention,” says Dr. DeBusk-Lane.

By addressing the issue of chronic absenteeism with advanced analytics, the conversation has now shifted for CCPS. Leadership is asking more questions about what data is available and how they are able to extract more value from it in other content areas. Building this first model opened up the door.

Dr. DeBusk-Lane says the response to the model has been encouraging. “There’s been a visceral and positive response. People have said, ‘If I had this last year, we could have doubled our accuracy rate for identifying students at risk of chronic absenteeism.’” His team built the model using machine learning on AWS with Amazon SageMaker. “Using Amazon SageMaker allowed me to quickly realign my efforts and optimize the actual process of training models,” says Dr. DeBusk-Lane.

Currently, the team is still working on the model—the next steps include digging deeper into the data, optimizing it, and working with content specialists and practitioners to look at the uncaptured variability to improve model accuracy. Dr. DeBusk-Lane says soon, his team hopes they will be able to iterate the model to the latest data and have more accurate, ready-to-use predictions. Once this model is optimized, he has plans to expand his portfolio for CCPS. “I’m interested in looking at how we can predict student performance and growth. This would really help teachers and division staff to tailor their approach based on student and class projections,” says DeBusk-Lane.

Read more education stories and machine learning stories on the AWS Public Sector Blog, including how Loudoun County Public Schools in Virginia digitally transformed using AWS. You can also learn more about how AWS is helping K12 education innovate faster through our Initiate eLearning hub.

Listen to the Fix This podcast to hear more education stories, including from Los Angeles Unified School District—the second largest K12 school district in the US—on the Mission Critical Cloud mini-series. You can stream all episodes on Spotify, Apple Podcasts, Google Play, Stitcher, TuneIn, Overcast, iHeartRadio, and via RSS.