Background. Diagnosis of early-stage chronic kidney disease (CKD) is a global challenge, as late-stage disease is more commonly diagnosed. The development of modeling methods for making management decisions aimed at improving the efficiency of early diagnosis of CKD is an important scientific and practical task, which can be greatly supported by the use of machine learning algorithms (MLA). Aim. To improve the accuracy of diagnosis of CKD using data from history, clinical-instrumental, genetic examination and machine learning algorithms (MLA). Methods. Data were obtained from a single-center retrospective catamnestic cohort study (2011–2022) of children with CKD stage 1–4 aged 1 to 17 years. The main group included 128 children with CKD, and the comparison group included 30 children without any kidney disease. Two groups were comparable by sex and age. The data of anamnesis, clinical-instrumental and genetic examination were used to build a model for CKD diagnosis. The model was built using the MLA multivariate logistic regression (MLR). Three variables were used in the model: erythrocyte sedimentation rate in blood (β = 0,392; p<0,001); erythrocytes in urine (β = 1,225; p<0,001); and asthenic physique (β = 5,792; p<0,001). Results. A diagnostics model was obtained allowing prediction of CKD on a test sample with accuracy of 90,3% [80,6; 96,8], sensitivity of 92,0% [81,5; 100,0], specificity of 83,3% [50,0; 100,0], ROC-AUC = 90,0% [77,2; 100,0]. The resulting model is of excellent quality (>90%) as the ROC-AUC is 0,90 on the test sample. The cut-off point value of the probability of CKD is 0,25. Conclusions. We developed and tested the model that diagnoses early-stage CKD in children with high accuracy