Increasingly commercial companies including Google, Amazon and Apple are using machine learning (ML) to predict customer behaviour and market trends. As these ML methods mature, they will continue to help improve commercial sector decision making, and potentially military processes as well. Reports suggest that the DoD alone could save $32 billion a year by increasing logistics and operational efficiency, savings that ML could help facilitate. Unfortunately, many ML methods require millions of known data points to train a system before its predictive capabilities can be realized. However, for many military processes, only relatively small data sets are available (i.e. hundreds to thousands of points). This paper explores a specific ML method, Bayesian Networks (BN), to function on problems with small amounts of known data. Specifically, this work investigates the feasibility of using Kriging and Radial Basis Functions to augment existing data available for training BNs. In addition, tuning BN parameters to increase network accuracy using Particle Swarm Optimization is also presented. Combined results from three different datasets suggest that pairing data generation and prior probability approximation can allow BNs to more accurately predict a system's outcome with small amounts of known data, potentially up to 80% or higher. Ultimately, as strategies outlined in the paper continue to develop they could help aid the implementation of BNs for a wide range of military processes. This would allow inefficiencies to be predicted before actual time, materials, and person hours are wasted.