Precision feeding aims to provide a ration close to the nutritional requirements of each gestating sow, which are calculated using a model (e.g. InraPorc) that requires input data (e.g. sow characteristics) and an estimate of the herd’s farrowing performance. This study evaluated machine-learning methods to predict the daily nutritional requirements of gestating sows based on sensor data as a function of different configurations of virtual farms. Data on 73 gestating sows were recorded by automatons or sensors such as electronic feeders, drinking stations, connected weight scales, and accelerometers. Nine machine-learning algorithms were then trained on various dataset scenarios as a function of different virtual farm configurations (using data from one or two sensors) to predict daily standardized ileal digestible lysine and metabolizable energy requirements for each sow. Adding the inputs usually provided to the InraPorc model (i.e. housing conditions and sow characteristics at artificial insemination) to the sensor data improved the mean average percentage error by 5.6 % for lysine and 2.2 % for energy. The highest coefficients of multiple determination for lysine (R2 = 0.99) and for energy (R2 = 0.95) were obtained for scenarios that involved an automatic feeder and included housing and sow characteristics. For these scenarios, the root mean square error was lower with Gradient Tree Boosting (0.91 MJ/d for energy and 0.08 g/d for lysine) than that with linear regression (2.75 MJ/d and 1.07 g/d, respectively). The results of this study show that the daily nutrient requirements of gestating sows can be predicted accurately with data provided by sensors and machine-learning methods, which paves the way to simpler solutions for precision feeding.