Determining the economic feasibility of a PV-based energy-producing system depends on accurate prediction of the performance and energy output of solar (PV) farms. Based on three environmental input values such as sun irradiation, wind speed, and ambient air temperature, this study endeavor to create a prognostic model for the annual power output yield of an installed PV system. Four machine learning methods including XGBoost, AdaBoost, linear regression (LR), and decision tree (DT) were employed to model and predict the power generation. The models were compared with statistical metrics like mean squared error (MSE), mean absolute percentage error (MAPE), and coefficient of determination (R2). With a training MSE of 1.218E + 07 and a training R2 of 0.98854, AdaBoost shows remarkable training accuracy, therefore suggesting its resilience in identifying fundamental patterns in the training data. During training phase, the decision tree model achieves 100% accuracy; yet it suffers from overfitting and shows notable prediction errors during testing. With a near-zero training MSE of 0.000683 and a flawless training R2 of 1, XGBoost stands out; nonetheless, on the test data, it performs favorably to other models with a model testing MSE of 2.66E + 08 and training R2 of 0.8412. With XGBoost’s lower test MAPE of 6.36 than AdaBoost's 5.966, XGBoost edges out and provides better training accuracy and fewer prediction errors.