Saving students with machine learning, data analytics, and some surprising insights
By Sarmila Basu, Aug 08, 2017
Sarmila Basu (second from left) meets with members of her Microsoft IT Data and Decision Sciences team.
The sun rises in the east, flowers bloom in the spring, and teachers inspire kids to do great things.
We know all this, right? Yes, but there is much more to it than that.
Thanks to data analytics and machine learning, we are now discovering that the exact words teachers use to give students feedback is among several factors that directly influence whether a student succeeds or fails academically. And furthermore, whether she stays in school or drops out.
At Microsoft, we set out to learn more about the causes behind high school students dropping out of school, so we partnered with the Tacoma School District in Washington state, which had one of the lowest graduation rates in the country (55 percent).
It all started when Shaun Taylor, their CIO, reached out to us. Like everyone in the district, he was frustrated with the high rate of failure, and he wanted to have a deep, authentic conversation about whether data could help. Shaun worked with my team—Microsoft IT Data and Decision Sciences—and Microsoft Consulting Services to first figure out which students were likely to drop out, and then determine what specific actions the district should take to keep each individual student in school.
Attendance rates, truancy, and discipline issues all contributed to dropout rates, but surprisingly, they were not the leading indicators that a child was entering the danger zone. According to our data investigation, unexpected catalysts had a bigger impact.
We found, for example, that written comments teachers left for at-risk students on their report cards and on tests had a major impact on whether the student eventually dropped out. If a teacher wrote, “excellent work, you are making great progress!” or “I see that you have great potential!” on a borderline student’s work, that student was much more likely to stay in school. If the teacher wrote, “this isn’t good enough, you can do better,” or “you need to work harder,” the chance of that student dropping out shot up considerably.
The biggest factor influencing whether a student was at risk of dropping out was how heavy their course load was. By studying the behind the district’s dropout record, our team successfully projected that students who had overly heavy classwork as juniors and seniors (often due to the retaking of courses they failed or dropped as freshmen and sophomores), were much more likely to leave school than their peers.
Thanks to predictive analytics, we could determine why and when a student would drop out. But that was not enough. We wanted to recommend preventive actions that the district could take to keep those students engaged and in school.
It was important for us to give the district individualized recommendations for each student projected to drop out, and it was even more important to have confidence that our recommendations would keep the kids in school (their futures were on the line!). From our findings, it became clear that teachers needed to be more positive in their feedback to students, and that the school district needed to limit the number of classes that students could take at once.
The school district was with us every step of the way, validating our findings, offering suggestions, implementing ideas, and keeping everyone moving. The district crafted an individualized outreach plan for each vulnerable student and then implanted it with care and determination.
And it worked!
In an amazing turnaround, by 2016 the school district graduation rate climbed to 83 percent — a couple of clicks above the national average — and far from the 55 percent number that had many calling it a “dropout factory.” The real story here are the people at the school district who were passionate about helping their students stay in school. Thanks to their persistence and the power of analytics, the district has totally turned itself around. Learn more about our efforts in Tacoma by checking out this case study.
We have since taken this model and are currently offering it to high schools across the country and around the world as part our education package. We have also begun experimenting at the college level, conducting research on risk factors older students face.
We invite you to come back soon to see those findings and hear other data analytics stories.
This blog post was originally published in Microsoft’s IT Showcase.
About the author
Dr. Sarmila Basu is Chief Data Scientist at Microsoft IT, leads the MSIT Data Sciences (DDSG) group and oversees both internal and external global customer engagements. Dr. Basu is instrumental in Microsoft’s forward-looking development efforts to integrate the company’s Data Science expertise with the Sales, Services and Product Group. In this role, she has been heavily influential in the direction and execution of the company’s commitment to leveraging Advanced Analytics solutions using Microsoft technology. Using this Data Science strategy maximizes management’s ability to make data driven decisions and solve challenging business problems.
Dr. Basu is a Ph.D. in Economics, and has 20 years of senior management experience working with Fortune 500 companies in Telecom, Financial Services and IT. She is passionate about Data Science and the business value it delivers. Sarmila firmly believes there is an art to Data Science. It requires more than Advanced Analytics depth and technical expertise to make a material difference in how we help our company and customers.
Dr. Basu is a noted and sought after conference speaker. Work from Dr. Basu’s team is featured on Microsoft’s IT Showcase. You can find Dr. Basu’s writings on Linkedin. When not at work, Sarmila is committed to her philanthropic pursuits. These interests include Economic Empowerment and as a board member for the Arts.
Accelerate your organization’s journey to analytics maturity
Get the data sheet to learn how the Research & Advisory Network advances analytics capabilities and improves performance.