Diagnostic accuracy remains one of the central problems of medical care. In this work we attempt to apply artificial intelligence to solve this challenge. We propose an approach to medical prediction based on the intelligent analysis of patients’ data from 200 different laboratory tests. The initial sample included 7, 918 cases falling into 4 nosological categories: D50 (iron deficiency anemia), E11 (non-insulin-dependent diabetes mellitus), E74 (other disorders of carbohydrate metabolism), and E78 (disorders of lipoprotein metabolism and other lipidemias), and was further divided into the training and testing datasets. Using gradient boosting, we constructed a machine learning model. The model demonstrated a recognition rate of 89 % (AUC-ROC) and a mean certainty in the diagnosis of 92 %. Our study proves feasibility of using machine learning in the analysis of this type of medical data. We are currently implementing a web-service for medical prediction as part of our Healthcare platform aiming at automation of clinical practice.