<p>Predicting phycocyanin (PC) and chlorophyll-a (Chl-a) concentration in lakes and reservoirs using machine learning and remote sensing is helpful. However, identification of optimum features and data imbalance during machine learning prediction development is limited. This study aims to investigate the prediction of PC and Chl-a concentration using Sentinel-2 and hybrid machine learning (ML) in Lake Ziway and Koka Reservoir. To accomplish this, a Python script was utilized to model the three single and three hybrid machine learning (ML) algorithms: support vector regression (SVR), back propagation neural network (BPNN), and random forest regression (RFR). This study used Cauchy and adaptive bald eagle search (CABES) for feature optimization and synthetic minority over-sampling technique (SMOTE) for data augmentation. The results of this study identified that the CABES-RFR performed better in predicting Chl-a and CABES-SVR performed better in predicting PC concentration in Lake Ziway and Koka Reservoir, with minimum root-mean-squared error (RMSE) and an efficient coefficient of determination (<i>R</i>²). The finding of this study shows that the Sentinel-2 data and hybrid ML are utilized as a tool in order to identify areas that are susceptible to eutrophication and coastal zones affecting lakes and reservoirs.</p>