In aquaculture, phytoplankton contribute to primary productivity, helping maintain fishery harvests and water stabilization. To better understand the drivers controlling phytoplankton diversity, sampling was conducted in 185 aquaculture ponds that cultured five species of aquatic animals (channel catfish, common carp, grass carp, crayfish and largemouth bass). Significantly higher α-diversity was found in the channel catfish and crayfish ponds than in the other ponds. Generalized linear model showed that phytoplankton density, cultivated animals, and conductivity can be used to explain phytoplankton diversity. Cultivated animals have an important influence on phytoplankton communities. Therefore, we suggest that in intensive fish-stocking ecosystems, top-down effect is an important strength on phytoplankton communities. Cyanobacteria blooms harm the harvesting of aquatic animals and threaten human health. It is thus environmentally beneficial to identify key drivers and develop methods for cyanobacteria bloom predicting. Based on monitoring data across 331 aquaculture ponds in central China, we developed two machine learning models - LASSO model and the Random Forest (RF) model - to identify key drivers and thereby predict cyanobacteria abundance. The Lasso model (R2 = 0.927, MSE = 0.310) outperformed the RF model (R2 = 0.567, MSE = 2.050) in the cyanobacteria abundance prediction. For well-equipped aquaculture ponds where water monitoring data are abundant, farmers can use the identified nine environmental variables by the LASSO model as an operational solution to accurately predict cyanobacteria abundance; while for crude ponds that have limited monitoring data, the three environmental variables identified by the RF model form a convenient solution for useful cyanobacteria prediction. Carbon reduction coupled with phosphorus reduction in feed usage may be a good management approach for cyanobacteria prevention toward a healthy ecological state in the aquaculture ponds.