<p>Water evaporation prediction for environmental monitoring, irrigation management, and meteorological modeling is growing nowadays. This study develops a new optimized deep learning model with a modified regression model for water evaporation prediction. Initially, in this study, a dataset was collected from publicly available sources on the internet. The collected dataset faced class imbalance issues; to address these issues Random Over-Sampling method (ROS) is utilized. Then, during preprocessing, Z-score normalization is employed to scale features to a specified range. Preprocessing is done to process the feature selection model, extended recursive feature elimination (EX-RFE) is used, which finds the important features and helps the effective prediction. Finally, they developed a novel optimized stacked long short-term modified support vector regression (OS-LSTM-MSVR) for effective water evaporation prediction, which enhanced performance and offered an optimal solution. Furthermore, a new enhanced Gauss map Frilled lizard optimization (GM-FLO) is utilized to fine-tune the hyperparameters of the developed model. Additionally, implementation outcomes demonstrate that the OS-LSTM-MSVR model obtained an MAE reduction of 0.3551 and an RMSE reduction of 0.5343 compared to the baseline approaches. The optimization significantly enhanced generalization, with an R² value of 0.9901 across all test folds.</p>