The curve magnitude of the deformed spine is the leading factor for treatment planning in patients with adolescent idiopathic scoliosis (AIS). Therapeutic intervention is typically recommended at a Cobb angle of 20°. While diagnosis traditionally relies on X-ray imaging, torsobarography offers a potential radiation-free alternative. This novel method analyzes the dorsal surface based on the pressure distribution in a supine position. From the pressure image, torsobarographic parameters are extracted to quantify the shape and symmetry of anatomically relevant structures. The objective of this study is to investigate the ability to differentiate the curve magnitude of AIS using torsobarographic parameters and machine learning algorithms. The following classifiers were examined: support vector machine (SVM), k-nearest neighbor (KNN), naive Bayes (NB), and decision tree (DT). The hyperparameter optimization was combined with a sequential forward floating selection (SFFS). Based on the cut-off values of the Cobb angle in the range of 10° to 20°, the dataset was divided into two groups. The variable cut-off values enable a systematic evaluation of model performance for early identification of AIS (Cobb angle < 20°). For a cut-off of 20°, an almost perfect model performance was achieved with a Cohen's kappa κ of 0.91, a sensitivity SEN of 0.92, and a specificity SPC of 0.98. However, even with lower cutoff values, a κ of more than 0.8 was realizable, indicating early AIS detection is potentially feasible with torsobarography. The generalizability verification revealed a decrease in model performance (κ > 0.6), achieving the best outcome at a cut-off of 19° with a mean κ of 0.79 ± 0.04, SEN of 0.88 ± 0.04, and SPC of 0.93 ± 0.02.Clinical Relevance— The findings of this study underline the potential of torsobarography for non-radiographic assessment of curve magnitude and early detection of AIS.