The article examines the optimization of the feature space of diagnostic indicators for conditions of hypo/hyperthyroidism. A correlation analysis was performed with the construction of a correlation matrix aimed at reducing the number of variables, as well as identifying strongly related pairs of diagnostic features involved in the diagnosis and treatment adjustment in people with diseases of hypo/hyperthyroidism. As a result, 13 pairs of strongly related features were identified. The study also used machine learning, and in particular the random forest method, which allowed for a meaningful analysis of the importance of diagnostic features, which helped determine which medical indicators have the greatest impact on diagnosis or prognosis when diagnosing and correcting treatment for hypo/hyperthyroidism; to identify patterns hidden in medical data. Within the framework of personalized medicine, the results showed that reducing the feature space will speed up diagnosis, increase the accuracy and interpretability of models describing conditions of hypo/hyperthyroidism, and identify subgroups of people with different symptomatic patterns. #COMESYSO1120