High-throughput phenotyping using remote sensors is a recent strategy that has enabled breeders to evaluate complex traits in breeding programs, such as internal CO₂ concentration (Ci). Ci is an important trait for measuring plant adaptation to climate change. However, this is only possible after validating models. The objective of this work was to identify promising machine learning (ML) models for predicting physiological traits in soybean genotypes using a hyperspectral sensor. The experiment was conducted in a randomized block design with four replicates and 25 soybean genotypes. Ci was evaluated 60 days after emergence of the genotypes in three leaves per plot. Spectral readings were performed on these same leaves with a spectroradiometer to acquire spectral variables in the range of 350 to 2500 nm. The spectral variables were used as input for the prediction of each physiological trait. The ML models tested were artificial neural networks, REPTree decision tree, M5P decision tree, random forest (RF), and support vector machine. A 10-fold cross-validation was used to obtain the following accuracy parameters: Pearson's correlation coefficient between the observed and predicted values, ​​and the mean absolute error. The results obtained indicate that all the models evaluated, with the exception of RPTree, are efficient for performing high-precision phenotyping of Ci using spectral variables as input. These results enabled the large-scale evaluation of soybean genotypes.