Machine Learning Models for Accurate Prediction of Obesity: A Data-Driven Approach

The number of people affected by obesity is rising steadily. Diagnosing obesity is crucial due to its harmful impacts on human health and it has become one of the world’s most important global health concerns. Therefore, it is crucial to develop methods that can enable early prediction of obesity risk and aid in mitigating the increasing prevalence of obesity. In the literature, some methods rely solely on Body Mass Index (BMI) for the prediction and classification of obesity may result in inaccurate outcomes. Additionally, more accurate predictions can be performed by developing machine learning models that incorporate additional factors such as individuals’ lifestyle and dietary habits, alongside height and weight used in BMI calculations. In this study, the potential of three different machine learning methods (naive Bayes, decision tree, and Random Forest (RF)) in predicting obesity levels were investigated. The best performance among the compared methods was obtained with RF (accuracy=0.8892, macro average F1-score=0.8618, Macro Average Precision (MAP)=0.8350, Macro Average Recall (MAR)=0.9122,). In addition, feature selection was also performed to determine the features that are significant for the estimation of the obesity level. According to the experimental results with feature selection, the RF method resulted in the highest score (accuracy=0.9236, MAP=0.9232, MAR=0.9358, macro average F1-score=0.9269) with fewer features. The results demonstrate that the performance of machine learning models on the same dataset can be enhanced through detailed hyperparameter tuning. Furthermore, applying feature selection can improve performance by mitigating the adverse effects of irrelevant or redundant features that may degrade the model’s effectiveness.