Today, 60 countries suffer from landmines and other unexploded ordnance, and more than 4,700 people were killed or injured by these explosives in 2022. At the same time, mine clearance is a lengthy and resource-intensive process. The areas to be searched for mines usually cover large regions, and skilled deminers, hardware, and funding are limited. Machine learning (ML) researchers have proposed several solutions to increase the efficiency and effectiveness of mine clearance. This brief review of the current research aims to provide an overview of findings on the use of machine learning to support ERW detection in clearance operations. In particular, it aims to answer which input features are used in research to train ML algorithms (RQ1), which ML algorithms are used to predict the presence of ERW (RQ2), and which gaps in research exist on this topic (RQ3). The review shows that studies on object detection, especially based on GPR and metal detector data, dominate the research while less than 10% of the research studies ML for ERW risk prediction. It also highlights the wide variety of input features used for ERW risk prediction algorithms. Furthermore, the review finds that SVMs and neural networks are the most studied ML algorithms to support ERW clearance. However, it also points out significant differences in ML techniques between the individual lines of research. Finally, the review highlights areas for future research. Most importantly, future research should focus on landmine risk prediction, work with real-world data and investigate performance improvements through input data fusion and incorporation of prior knowledge such as ERW patterns.