Applying Predictive Analytics in Assessing Health Conditions of Applicants
Abstract
Predicting military attrition due to conditions that existed prior to service is a complicated problem. My thesis explores underwriting practices and risk assessment in the life and health insurance industries with the aim to link private sector underwriting techniques to the military medical screening process. I review the current prediction models in the economic, actuary, and medical fields and find many of these models utilize complicated machine-learning algorithms to include random forests, deep convolutional neural networks, and deep dynamic memory neural network models. For my empirical analysis, I utilize a Cox proportional hazard model to determine risk via potential predictor variables. My findings suggest past self-inflicted injuries, substance use disorder (current and in the past), waivers for drug offenses, missing an Armed Forces Qualification Test (AFQT) score, and deployments (current and in the past) are associated with higher hazard rates of separation. This information provides insights regarding the separation risks associated with various indicators.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 2021
- Accession Number
- AD1150674
Entities
People
- Mark A. Knutson
Organizations
- Naval Postgraduate School