Using machine learning to predict obesity in high school students

Four enhanced machine learning models were used to predict obesity in high school students by focusing on both risk and protective factors: binary logistic regression; improved decision tree (IDT); weighted k-nearest neighbor (KNN); and artificial neural network (ANN). Nine health-related behaviors from the 2015 Youth Risk Behavior Surveillance System (YRBSS) for the state of Tennessee were used as model inputs. Results show that, compared to the logistic regression model that achieved 56.02% accuracy and 54.77% specificity, IDT, weighted KNN, and ANN each performed significantly better. The IDT model achieved 80.23% accuracy and 90.74% specificity, while the weighted KNN model achieved 88.82% accuracy and 93.44% specificity. The ANN model achieved 84.22% accuracy and 99.46% specificity. Implications and suggestions for slowing the increase in adolescent obesity are discussed.