The detection of missing radioactive material within a particular sample is advantageous for various applications of Materials Accountancy and Non-Destructive Assay. Examples of these applications include the monitoring of spent fuel pools and casks, as well as the inspection of fresh fuel assemblies. Currently employed methods for these processes include the use of passive or active gamma ray detection, where variation in detector responses are used to deduce if there is missing material and determine its expected location. This work investigates the feasibility of using machine learning algorithms for processing detection data in these scenarios to improve overall sensitivity. Preliminary simulated trials with a grid of nine 137Cs point sources and two NaI detectors show that a k-nearest neighbor algorithm can successfully predict the location of a missing source with 100% accuracy. Similar preliminary trials with up to two missing sources yielded an accuracy of 99%, suggesting that machine learning has promise for this application. These initial studies, as well as results with larger grids of sources, and trials with measurements taken in a laboratory setting are included.