In recent years, machine learning has evolved rapidly as arithmetic power has increased, algorithms have been updated, and data has expanded. It not only provides new tools and methods for ecological research, but also promotes the updating of research ideas and the transformation of research pattern, leading ecology gradually to the era of data-driven research. Meanwhile, ecological security is a research hotspot in ecology. With the increasingly serious environmental problems, ecological security has gradually become the focus of global attention. In China, ecological security is also highly valued by the government. Ecological security includes three subfields: ecological security assessment, modelling and security early warning, and ecological security patterns. In this paper, we first outline the content, research methodology, and shortcomings of the three subfields in ecological security. Then we systematically sort out the main applications of machine learning in the field of ecological security, and summarize the keywords and related algorithms, based on the core collection of Web of Science and CNKI. The results show that the main applications of machine learning in ecological security assessment include 1) analyzing and determining influencing factors and 2) analyzing and determining the weights and importance of indicators. Applications in modeling and security early warning include 1) modeling the correspondence between ecological security and indicators and 2) dynamic prediction and early warning. And applications in ecological security patterns include 1) identifying ecological sources (zones) and 2) the spatial and temporal evolution of patterns. Then we deeply analyze the common algorithms within each application, as well as the characteristics, limitations and applicability of the algorithms in the field of ecological security. And we summarize the evolutionary patterns of machine learning algorithms in ecological security, including 1) optimization of the algorithms themselves, 2) inter-combination of different algorithms (integrated learning algorithms) and 3) multiple possible algorithms for solving the same problem. The challenges faced by current research include the high threshold of technical implementation, the precision and rigor requirements of application scenarios, the lack of interpretability of some models, and the limited generalization capability. Future research directions where machine learning and ecological security are expected to be deeply integrated are focused on 1) identifying key ecological safety thresholds, 2) constructing efficient and accurate ecological safety early warning platforms, and 3) elucidating the intrinsic driving mechanisms of ecological pattern evolution. It provides valuable references and insights for future exploration in the field of ecological security, promotes interdisciplinary cooperation, and responds to the complex and changing ecological security challenges with technological innovation.