In geothermal developments, characterizing hydrothermal flow is essential for predicting future production and designing effective development strategies. Numerical simulation models require determining a large number of input parameters to represent a reservoir. Most previous methods have estimated plausible parameters through a trial-and-error search with measurement data, which is time-consuming and dependent on the subjectivity of the analyst. In this study, we propose a machine-learning-based method to estimate input parameters (i.e., permeability distributions, heat source, and sink conditions) for geothermal reservoir modeling. A large amount of training data was prepared by a geothermal reservoir simulator capable of calculating pressure and temperature distributions in the natural state. Machine learning algorithms were applied to classification and regression problems, and Gradient Boosting Machine was ultimately selected. The performance evaluation scores were high for estimating parameters even for the 3D problem. The machine learning model was applied to three-dimensional data simulating a geothermal field. Although the training data had a different distribution, the estimated distribution of permeability showed the same trend as the true distribution. Recalculation using the estimated input parameters resulted in temperature distributions with good accuracy. This study successfully demonstrates the possibility of applying the model to field data.