Machine Learning, Time Series, and Survival Analysis for Describing Substance Abuse and Relapse Among U.S. Navy Personnel
Abstract
The U.S. Navy (USN) has targeted drug and alcohol abuse within its ranks by means of screening, training, and treatment; however, substance abuse remains a threat to the mission and to the personal health and well-being of USN service members. Therefore, we explore new methodology to identify detrimental behaviors associated with relapse and to forecast substance abuse occurrences. Furthermore, survival times for relapses are assessed to inform treatment program enrollment timelines. This research uses the USN Alcohol and Drug Management Information Tracking System (ADMITS) data and classification, time series, and survival analysis machine-learning models to address our research objectives. We apply performance measures to assess a models accuracy and goodness of fit to distinguish the top-performing models and their consideration for subsequent use. Additionally, we perform supplementary analysis on U.S. duty location, USN rank, and USN occupation subgroups to identify over- or under-representation over time in terms of substance abuse occurrences. Our results identified service members with prior relapses to be of most importance in predicting future relapse, strong seasonality in alcohol abuse occurrences, and a subset of USN service members who relapse faster than observed treatment timelines. Additionally, many U.S. duty location, USN rank, and USN occupation subgroups were discovered to be trending above or below the baseline proportion of alcohol abuse occurrences.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2022
- Accession Number
- AD1185044
Entities
People
- Chad J. Minnick
Organizations
- Naval Postgraduate School