<![CDATA[Obesity is a growing global public health concern that increases the risk of chronic diseases and significantly affects quality of life. Traditional diagnostic methods such as Body Mass Index (BMI) have limitations in accurately representing body fat distribution and individual health conditions. This study aims to comparatively evaluate the performance of various machine learning and neural network models in predicting obesity levels using a multiclass classification approach. The dataset consists of 2,111 observations with 12 predictor variables and seven obesity categories, obtained from a publicly available source. Data preprocessing included duplicate removal, outlier handling using the interquartile range method, feature scaling, and categorical encoding, followed by a 60:20:20 train–validation–test split. Several classifiers were implemented, including Logistic Regression, Support Vector Classifier, Random Forest, Extra Trees, Gradient Boosting-based models (XGBoost and LightGBM), Multilayer Perceptron, K-Nearest Neighbors, and TabNet. Model performance was evaluated using macro-average F1-score and confusion matrix analysis. The results indicate that LightGBM achieved the highest predictive performance with an F1-score of 0.96, demonstrating strong generalization across obesity categories. XGBoost and Random Forest also showed strong performance, while Support Vector Classifier exhibited consistent results across training, validation, and cross-validation. These findings suggest that ensemble-based models are highly effective for obesity classification, while model selection should consider accuracy, interpretability, and computational constraints.]]>
Copyrights © 2026
