Forecasting enlisted attrition in the United States Marine Corps by grade and years of service

Abstract

The purpose of this thesis is to analyze historical United States Marine Corps enlisted attrition behavior and apply time series forecasting techniques by grade and Years of Service in order to identify methods to improve manpower analysts' ability to effectively forecast attrition behavior. This study compared the results of one to five-year Moving Average models and the results of one to five-year Weighted Moving Average models based on two Measures of Effectiveness, Mean Square Error and the Mean Absolute Percent Error. The results of the Friedman test indicate statistical significance of the results in relation to the Mean Square Error of the one to two-year Moving Average models. This thesis demonstrates that in most cases, a simple one-year Moving Average more effectively estimates attrition behavior than the other Moving Average or Weighted Moving Average models analyzed.