Aiming at the problems of sparse data, strong nonlinearity and insufficient mechanism characterization in aquatic ecological environment research, this paper uses literature published between 2014 and 2024 retrieved from the China National Knowledge Infrastructure (CNKI) and the Web of Science Core Collection. On the basis of sorting out the basic principles of classical machine learning, deep learning, ensemble learning and automated machine learning, this paper reviews the application progress of machine learning in the field of aquatic ecological environment from four research scenarios, including water quality prediction and management, water pollutant analysis, eutrophication and cyanobacterial bloom research, and aquatic ecosystem protection. Furthermore, a comparative analysis is carried out based on data characteristics, task objectives and model performance, and problem-oriented suggestions for algorithm selection are proposed. The results indicate that ensemble learning algorithms such as random forest and XGBoost have good applicability for small- and medium-sized samples, the mining of multi-factor nonlinear relationships and the identification of variable importance. Models including LSTM and GRU are more applicable to multi-site and multi-temporal scale sequence prediction. CNN, Transformer and other models show considerable application potential in remote sensing recognition and algal biological image analysis. At present, the application of machine learning still faces several challenges, including inconsistent data quality, insufficient model interpretability, high computational cost, and inadequate collaboration with mechanistic models. In the future, by leveraging data-sharing systems, the application scope of machine learning in the aquatic ecological environment field can be further expanded.