In this paper, we propose an approach CropWaterNeed in order to estimate and predict the future water needs and maximize the productivity in the irrigated areas. Unfortunately, we have not identified data available to be employed in such machine learning process in order to predict plants water needs. The proposed approach consists of extending the classic machine learning process. Particularly, we define a process to build dataset that contains plant water requirements. To collect data, we extract meteorological data from Climwat database and plants water requirements using Cropwat Tool. Then, we aggregate the extracted data into a dataset. Subsequently, we use the dataset to perform the learning process using XGBRegressor, Decision Tree, Random Forest and Gradiant Boost Regressor. Afterward, we evaluate the model generated by each algorithm by measuring the performance measures such as MSE, RMSE and MAE. Our work shows that the model generated by XGBRegressor is the most efficient in our case while Random Forest is the least efficient. As future work, we aim to apply the proposed process to test the performance of other regression algorithms and to test the impact of using deep learning techniques with the extracted data.