EVALUATION OF MACHINE LEARNING APPLICABILITY FOR USMC REENLISTMENT

This research examines the applicability of machine learning algorithms to best predict the probability of reenlistment of enlisted first-term Marines. Given the availability of data today, machine learning can be a useful tool to make policy decisions that can impact the future Fleet Marine Force. This thesis uses demographic data, pre-boot-camp data, performance indicators, legal data, awards data, and selective reenlistment bonus indicators to identify factors that contribute to the prediction of reenlistment. This thesis applies data from the Total Force Data Warehouse (TFDW) and fits machine learning algorithms to assess their prediction accuracy. Measuring machine learning models by accuracy alone is not sufficient. An evaluation of top predictors is conducted to choose the best-preforming machine learning algorithm. Given the data used in this thesis, the machine learning algorithm that best predicts the probability of reenlistment is the C5 algorithm. Variables associated with deployment and performance are among the top ten predictors of importance.http://archive.org/details/evaluationofmach1094564888Captain, United States Marine CorpsApproved for public release; distribution is unlimited