CLASSIFICAÇÃO DO PERCENTUAL DE GORDURA CORPORAL EM ADULTOS HÍGIDOS E COM ALTERAÇÃO NA FUNÇÃO RENAL

Abstract

In the current health scenario, the epidemic of overweight and obesity has constantly increased. Excess body weight contributes to the development of several pathologies, such as cardiovascular diseases (CVD), diabetes and chronic kidney disease (CKD). CKD is a global public health problem due to its high morbidity and mortality. Machine learning techniques are of great value for data analysis in the healthcare sector. The aim is to develop a system that classifies body fat percentage using non-invasive, low-cost indicators in patients with and without chronic kidney disease. This is a cross-sectional study, in which the sample consisted of individuals treated at the Center for the Prevention of Chronic Kidney Diseases, made up of participants of both genders, aged between 18 and 78 years. Data were collected through a semi-structured questionnaire for sociodemographic and lifestyle data, and anthropometric measurements. For the implementation of the classifier and metrics, the performance of the classifier algorithms was used to analyze the performance measures of accuracy, sensitivity and specificity. The Android Studio development environment was used and the software was evaluated through usability testing with healthcare professionals and users from the general population. The present study was approved by the Ethics and Research Committee (CEP) of the Federal University do Maranhão CAAE: 67030517.5.0000.5087. A total of 244 individuals were obtained, of which 76.2% (n=186) were female and 23.8% (n= 58) were male. The average age of the sample was 46.14 ± 15.19 years. Decision tree presented itself as the most appropriate mathematical model with greater sensitivity and specificity. From testing the Decision Tree model, the variables with the best performance for classifying fat percentage were: gender, age, weight, height, waist circumference, arm circumference, calf circumference, SBP and DBP. The computational model used in this study showed excellent performance in classifying body fat percentage in adults using non-invasive variables.