We’re starting to learn more and more how data can change the way we live and improve the services that we rely on. Whether it’s predicting the next water main to break in the city of Syracuse or using predictive analytics to improve emergency medical services in Cincinnati, Data Science for Social Good — a grant-funded summer program that trains selected fellows in the implementation of data science projects at the University of Chicago — is helping pave the way for data to change the way cities operate.
“The goal of our efforts is to see governments [and] nonprofits use data to improve what they do,” said DSSG Director Rayid Ghani at the group’s annual conference in Chicago last month. “Data is not going to give them all the answers, but it will help them improve.”
For the past three years, DSSG has been seeking out and accepting applications from nonprofits and local governments around the country and beyond that have a social problem in need of addressing and the data available for data science fellows to work with. DSSG matches them with teams of recent data science graduates who dedicate three months to developing a solution to the problem. At the program’s annual gathering in August, DSSG 2016 fellows demonstrated their ability to improve critical government services, leading workshops on best practices and presenting the results of this year’s projects. Teams of DSSG fellows were stationed for the summer in cities and counties from New York to Kansas and Sedesol, Mexico and implemented machine learning and data-driven models to improve government services.
Models for Identifying Students in Need
In the Midwest, two groups of DSSG fellows teamed with school districts and other departments to improve the lives of students. Milwaukee, Wisconsin has an alarming rate of juvenile arrests — in 2012 the national average was 3,000, but Milwaukee’s rate was four times that. In an effort to combat this social problem, DSSG fellows created a predictive model using education and criminal justice data available through Milwaukee Data Share to identify students at risk of interacting with the juvenile justice system. While the current method for identifying students in need of intervention flags 22,000 students in the district and correctly flags 54 percent, the DSSG model flags only 12,000 with a 66 percent success rate, allowing for more accurate and effective resource allocation.DSSG fellows working with the Tulsa Public School system in Oklahoma developed a predictive model that puts information about which students are at risk of not finishing high school in the hands of teachers as early as the third grade. Although the amount of educational data at the third grade level is minimal, DSSG fellows used data from intermediate test scores, the amount of time students spent using educational software, and whether they attended after-school programming to build a predictive model with a dashboard that shows teachers which students need help the most.
The DSSG model in Tulsa identifies 250 more students per year than the school system’s prior strategy and ensures that 95 percent of the students identified as at-risk receive the help they need. It also includes a dashboard for teachers that not only shows whether a student needs additional help, but recommends specific strategies — summer school, learning apps, or more time with certain software — to help the student get on the right educational track.
Aiding the Efforts of Police to Avoid Negative Interactions with Citizens
Amidst waves of unrest over cases of use of force by police across the country, DSSG fellows set off to Nashville, Tennessee to help one department better allocate its resources to prevent adverse officer-citizen interactions before they occur. Using Nashville police data about past use of force incidents labelled improper by the department, accidents or injuries deemed preventable, and filed complaints that led to disciplinary action, the fellows set out to create a predictive model that would identify officers at risk for an adverse interaction.The fellows combined the incident and complaint data with departmental data about individual officers’ characteristics, behavioral history, and work history to identify high-risk officers and suggest intervention by department officials. This model proves far more effective than traditional, blanket intervention which identifies two thirds of the department for training. The DSSG model flags less than half of the number flagged by the traditional model, proving it to be far more useful; according to the fellows’ report, only five percent of officers in Nashville are involved in adverse interactions annually.
Helping Keep the Mentally Ill Out of Prison
Jails are proven to be a costly and far-from-ideal solution to addressing mental health issues in communities. In an effort to help avoid the incarceration of mentally ill citizens, DSSG fellows worked with EMS providers, mental health services, and the criminal justice system in Johnson County, Kansas. The fellows pulled historical data from the three departments and combined the data with demographic information and individual risk factors to feed a machine learning model that output scores assessing the risk for every individual in the county of entering prison.Ideally, the risk scores will be put in the hands of mental health workers that are already embedded with police in the county. This way, these mental health workers and their police partners can prioritize their outreach, potentially preventing a mentally ill individual from experiencing a crisis that would have previously led to crime and incarceration.
