Rastreamento do excesso de gordura corporal em adolescentes através de técnicas de aprendizado de máquina

• Autor(a) principal: SOUSA, Nilviane Pires Silva
• Data de Publicação: 2018
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da UFMA
• Texto Completo: https://tedebc.ufma.br/jspui/handle/tede/tede/2621
• topic: Composição corporal; Aprendizado de máquina; Programas de rastreamento; Sensibilidade e especificidade; Body composition; Machine learning; Mass screening; Sensitivity and specificity; Nutrição; Teoria da Computação

In the last two decades several developing countries have undergone an accelerated nutritional and epidemiological transition, causing an increase in the prevalence of excess body fat in adolescence in these countries, including Brazil. The high prevalence of overweight in this phase is associated with the early development of several diseases including metabolic and cardiovascular disorders, therefore, low cost screening methods are essential for the screening of excess general adiposity in this age group. Thus, the present study aims to classify excess body fat in schoolchildren using machine learning methods. Thereunto, three methods of classification were tested: k-nearest neighbors, support vector machine and decision tree. This is a cross-sectional study, where the database used for the training and test stages was collected in schools of the public system of São Luís / Maranhão, in the year 2018, consisting of 602 adolescents, of both genders, with age from 10 to 19 years. For external validation of the algorithm, another database of 320 adolescents, also from the school population, was used. A priori, the following indicators were evaluated: body mass, height, age, gender, waist circumference, hip, neck, calf and arm, heart rate, body fat percentage, body mass index and waist height ratio. For the application of the classifier algorithm and software development, the MATLAB® program was used, and the SPSS® software was used in the statistical analysis. The following statistical tests were applied: Kolmogorov-Smirnov, Student's T, ANOVA One Way, Mann-Whitney U and Kruskal-Wallis H. The classifier used in the construction of the software was the nearest k-neighbors that obtained accuracy of 78%, sensitivity 92% and specificity 54%, using the following entries: body mass, height, age, gender and waist circumference. When compared to body mass index and waist height ratio, the k-nearest neighbors achieved better performance (sensitivity 95%) in the screening of adolescents with high body fat percentage. Thus, the k-nearest neighbors algorithm can be used as a screening method with high sensitivity and low cost in the evaluation of general adiposity in adolescents from São Luís/MA.