COMPARISON OF CLASSIFIERS FOR THE RISK OF OBESITY PREDICTION AMONG HIGH SCHOOL STUDENTS

Obesity, which negatively affects human health, is a chronic disease due to genetic and living conditions. In this study, it was aimed to examine the observations with three main techniques: logistic regression, artificial neural networks and Naive Bayes, where the response variable was two categories of obese/not obese. Obesity questionnaire data, that was answered by 504 senior students in three randomly selected high schools in Gaziemir, Izmir, were analysed, and the predictive competences of the results of the three methods were evaluated. It was found that obesity is affected by the mother and father’s being obese and eating too much fruit. In addition, gender and diet status were significantly related with the obesity risk.   In the artificial neural network, backward propagation learning algorithm was used as the learning rule in the adjustment of the connection weights according to the output. With the Naive Bayes method, a classification based on the probability values ​​of the data was performed. The logistic regression model coefficient values ​​were determined, using the maximum likelihood method. According to obesity questionnaire data, it was determined whether the relationship of each obesity risk factor with the response variable was statistically significant. The Naive Bayes method has the highest accuracy in prediction obesity compared to the other two methods.