Aim. To develop and validate in silico models for predicting the antidiabetic activity of natural compounds and to test their performance on representative medicinal plant components. Materials and Methods. Two machine learning models based on the XGBoost and LightGBM algorithms were constructed and verified using a set of compounds from traditional Chinese medicine (TCM) formulations with experimentally confirmed hypoglycemic activity. The validated models were subsequently applied to analyze components of Galega officinalis L. (non-alkaloid fraction), whose composition was determined by gas and liquid chromatography-mass spectrometry (GC-MS and LC-MS). Results. The constructed models achieved an accuracy of 80-81% and were verified by correctly identifying active compounds among those known from TCM formulations to be effective in type 2 diabetes management. This confirms their ability to accurately classify bioactive natural substances. The models were applied to the components of the NA extract of G. officinalis L., a promising plant for subsequent studies on antidiabetic effects. Conclusions. The developed in silico models enable the prediction of antidiabetic activity of naturally derived compounds. Their verification on reference compound sets and application to G. officinalis L. extract demonstrate the potential of this approach for identifying low-toxicity bioactive substances.