Enhancing Obesity Risk Classification: Tackling Data Imbalance with SMOTE and Deep Learning

Data imbalance is a significant challenge in classification models, often leading to suboptimal performance, especially for minority classes. This study explores the effectiveness of the Synthetic Minority Over-sampling Technique (SMOTE) in improving classification model performance by balancing data distribution. The evaluation was conducted using a confusion matrix to measure prediction accuracy for each class. The results indicate that SMOTE successfully enhances minority class representation and improves prediction balance, although some misclassifications remain. Therefore, in addition to oversampling, additional approaches such as class weighting or ensemble learning are required to further improve model accuracy. This study provides deeper insights into the role of SMOTE in addressing data imbalance and its impact on classification model performance.