Stacking Ensemble Based Hybrid Machine Learning Approach for Predicting Obesity Levels

This research aimed to predict obesity levels using hybrid machine learning (ML). A dataset containing 2111 records and 17 features was used in the study. The Stacking Ensemble method was applied, which combines five basic ML algorithms (Random Forest, Extra Trees, Gradient Boosting, Linear SVM, and Naive Bayes). In the model where Multi-Layer Perceptron (MLP) was used as a meta-learner, obesity risk classes were tried to be predicted based on physical status and nutritional habits. The proposed hybrid model's prediction accuracy rate was 98.58%. This rate has a high accuracy rate compared to other studies. In contrast, the reliability and discriminatory power of the model were proven with a Kappa value of 0.9834 and an AUC score of 0.9917. In addition, high F1-Score (%98.59), Precision (%98.63), and Recall (%98.58) values reveal the consistency of the classification performance of the model. Among the basic models, the Support Vector Machine (SVM) performed best with an accurate rate of 96.69%. This model, which achieved a higher accuracy rate than similar studies in literature, can provide decision support to health experts and policymakers on obesity. The research demonstrates the effectiveness of machine learning techniques in obesity risk classification and offers promising results for future health applications.