Predictive obesity among children and adolescent using logistic regression and random forest model

Childhood obesity has become a serious public health concern in Malaysia and worldwide. This study aims to predict obesity among children and adolescents using two classification techniques: logistic regression and random forest. The dataset was obtained from Kaggle and includes 2,111 records with 16 demographic, dietary, and lifestyle variables. Logistic regression was used to identify significant predictors, while the random forest model captured complex interactions among variables. The performance of both models was evaluated using accuracy, precision, recall, specificity, F1-score, and AUC. The logistic regression model achieved high recall (98.9%) but very low precision (2.0%) and F1-score (3.8%), indicating a high false-positive rate. The random forest model showed more balanced metrics with a precision of 48.9% and F1-score of 46.1%, although the accuracy was lower at 49.9%. The findings suggest that while logistic regression can identify at-risk individuals, random forest provides more reliable predictions. Both models, however, showed weak performance overall (AUC ≈ 0.5), highlighting the need for improved data and modeling approaches