By: Eric Siegel, Program Chair, Predictive Analytics World
In anticipation of her upcoming keynote address at Predictive Analytics World Las Vegas, June 16-20, 2019, we asked Dyann Daley MD, Founder and CEO at Predict-Align-Prevent, a few questions about incorporating predictive analytics. Catch a glimpse of her presentation, Finding At-Risk Children with Geospatial Machine Learning, and see what’s in store at the PAW conference in Las Vegas.
Q: In your work with predictive analytics, what behavior or outcome do your models predict?
A: Child abuse and neglect
Q: How does predictive analytics deliver value at your organization – what is one specific way in which it actively drives decisions or operations?
A: Children age 0-3 years are most vulnerable to child abuse and neglect and also most likely to die from it. Almost half of children who die of child maltreatment are not known to child welfare agencies. We use place-based predictive analytics to find these invisible children so resources can be directed toward prevention.
Q: Can you describe a quantitative result, such as the predictive lift of your model or the ROI of an analytics initiative?
A: Two main goals of national child maltreatment prevention initiatives include reducing the number of children removed from home and placed into foster care and reducing the number of maltreatment deaths. To achieve these goals, first, we have to find the most vulnerable children. The case we will present at PAW demonstrates the following.
The predictions of the meta-model were classified into five risk categories and compared to a kernel density estimation. The meta-model outperformed the KDE in the highest risk category, capturing 70.73% of recorded maltreatment events, whereas KDE captured only 33.64%, thus demonstrating that our methodology more accurately predicts child maltreatment than the traditional spatial interpolation approach being used today.
These risk levels were used to determine whether removals of a child from home occurred in areas of high risk of child maltreatment. 64.13% of removals were located in areas classified as the highest risk category. The fourth risk level saw 26.36% of removals, 7.07% of removals fell within the third risk level, and 0 .5% and 1.9% of removals occurred in the two lowest risk categories, respectively,
Upon calculating the percent of child maltreatment fatalities that occurred in each risk category, we found no child maltreatment fatalities in the bottom three risk quintiles. Rather, 41.7% of child maltreatment deaths occurred in the fourth risk level, and 58.3% of child maltreatment fatalities were located in the highest risk category.
Q: What surprising discovery or insight have you unearthed in your data?
A: Domestic violence is the most predictive feature for all types of child maltreatment across space. Additional important features are more related to Adverse Childhood Experiences* (ACEs) than they are to poverty alone. Twenty years of research shows that exposure to four or more ACEs has an indelible negative effect on the physical, mental, educational, social, and economic futures of the majority of exposed people. ACEs events themselves, and the outcomes exposure produces, are the most useful predictors.
*ACEs include: Physical abuse, Sexual abuse, Emotional abuse, Physical neglect, Emotional neglect, Mother treated violently, Household substance abuse, Household mental illness, Parental separation or divorce, and Incarcerated household member.
Q: Sneak preview: Please tell us a take-away that you will provide during your talk at Predictive Analytics World.
A: One of the most significant threats to our population’s health and prosperity is child abuse and neglect. Place-based predictive analytics can provide tactical intelligence for community-based prevention.
Don’t miss Dyann’s keynote address, Finding At-Risk Children with Geospatial Machine Learning, at Predictive Analytics World on Tuesday, June 18, 2019 from 9:15 to 9:35 AM. Click here to register for attendance.
By: Eric Siegel, Conference Chair, Predictive Analytics World