Predictive Modeling for Early Identification of Suicidal Thinking in Social Networks
Abstract
One of the fundamental questions facing social science is how social networks and the cognitions people have about their networks impact their mental states and mental health. For many populations, these processes can be a matter of life and death. Suicide rates among active duty service members and homeless youth both dramatically outpace those seen many other populations. While there is a wealth of research across populations demonstrating the associations between social support and mental health, more sophisticated theorizing and modeling is needed to understand how the networks which surround individuals and the messages transmitted across those network ties interact over lime to engender mental states and mental health. How network dynamics impact person s mental health can often be seen most acutely during times of stress and transition. Two seemingly disparate populations facing such stresses are active duty service members and homeless youth. Major military transitions include joining and leaving the military, moving to a different duty assignments in the US and abroad and preparing and returning from deployments. It has been hypothesized that military transitions arc inherently stressful because transitions produce significant disruptions in the service member s social network. From a theoretical perspective, however two general classes of network ties are critical to this process, ties within a service member s unit and the disparate set of tics outside a unit (i.e. family members, friends outside of the service etc.). For homeless youth, a similar dichotomy or network influences impact mental health. Ties to other street youth are critical, but so too are ties to people outside of street life, such as family members, friends from home, and social service providers. For example, homeless youth are more likely to report suicidal thinking when they are connected to a greater number of depressed youth within a given street network. In contrast, youth who are more connected to friends from home and family are less likely to report depression and anxiety symptoms. Social science approaches to studying the linkages between mental health and social network dynamics can be greatly augmented by a close collaboration with computer science, and specifically AI. We need predictive analytics" to indicate onset of mental health issues based on specific network dynamic patterns. In the past, we have used dynamic bayesian networks, gaussian mixture models, ensemble decision trees and others to infer spatio-temporal patterns in modeling human behavior in data from real-world domains (e.g., predicting poaching attacks in national parks in Africa or in crime patterns in urban settings such as Los Angeles), as well as in data from lab experiments (e.g., on network games played by human subjects). These techniques were tested with real-world data on urban or wildlife crime as well as laboratory data, and shown to provide more accurate predictions than other competing techniques in multiple spatio temporal domains. We propose to leverage and build on these techniques, possibly enhanced with recent work on link-prediction, for this current proposal. To employ AI techniques to understand and model network influence, this proposal focuses on addressing two core challenges. In particular, we will conduct research in: (i) machine learning and predictive modeling of mental health (suicidal thinking and depression) and its connections to social network connections among homeless youth over time, using extant data collected previously by the team; and (ii) analogous machine learning and predictive modeling of mental health (suicidal thinking, post traumatic stress, and depression) and its connection to social networks over time among active duty service members, using new data to be collected from 200 soldiers who will be recruited through 3rd Infantry Division at Fort Steward, GA who are preparing for a deployment.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Oct 11, 2018
- Source ID
- W911NF1710445
Entities
People
- Eric Rice
Organizations
- Army Contracting Command
- United States Army
- University of Southern California