Identifying Risk and Protective Factors for Attrition Among Recently Enlisted Navy Personnel Using Variable...

Area of Science:

• Military medicine
• Data science
• Personnel management

Background:

• Navy recruit attrition impacts force readiness, with approximately 25% not completing obligated service.
• Identifying attrition predictors is crucial for retaining skilled personnel and optimizing military effectiveness.

Purpose of the Study:

• To identify key factors contributing to Navy attrition using advanced machine learning techniques.
• To compare the predictive performance of machine learning models against traditional regression analyses.

Main Methods:

• Utilized random forest (RF) and extreme gradient boosting (XGBoost) models on a dataset of 39,866 Navy personnel from 2016.
• Incorporated 542 independent variables, including demographics, medical visits, and medications, to predict attrition (discharge before end of service).
• Assessed variable importance measures (VIM) to identify significant predictors of attrition.

Main Results:

• RF model achieved the highest accuracy (81.7%) and area under the curve (90.0%), outperforming logistic regression.
• Top predictive factors for attrition included mental health disorders (e.g., adjustment disorders), specific occupations (e.g., Seaman Specialists), and demographic/sex-related issues.
• Five key predictive groups identified: mental health, occupations, demographics/sex, pain management, and medical compliance.

Conclusions:

• Machine learning, particularly RF and XGBoost, offers superior predictive power for Navy attrition compared to traditional regression.
• Variable importance measures effectively highlight critical factors influencing attrition, aiding in personnel management and retention strategies.
• Ensemble machine learning approaches provide robust, accurate, and reliable models for predicting military personnel attrition.