|
Obesity level prediction using deep learning approach – A comparative analysis |
|---|---|
| รหัสดีโอไอ | |
| Creator | 1. Srinivasa Gupta Nagarajan 2. Valarmathi Balasubramanian 3. Phani Gonugunta 4. Saran Kumar Gudla |
| Title | Obesity level prediction using deep learning approach – A comparative analysis |
| Publisher | Faculty of Engineering, Khon Kaen University |
| Publication Year | 2567 |
| Journal Title | Engineering and Applied Science Research |
| Journal Vol. | 51 |
| Journal No. | 4 |
| Page no. | 540-554 |
| Keyword | Obesity level, SMOTE, TabNet Classifier, XGB Classifier, GB Classifier |
| URL Website | https://ph01.tci-thaijo.org/index.php/easr/index |
| Website title | Engineering and Applied Science Research |
| ISSN | 2539-6161 |
| Abstract | Obesity, the excessive accumulation of body fat, affected millions globally and was influenced by eating habits, lack of activity, genetics, environmental factors, and emotional strain. It could lead to severe health issues, including insulin resistance, cardiovascular diseases, cancer, sleep apnea, joint problems, and mental health disorders. This study aimed to predict obesity levels using Machine Learning (ML) and Deep Learning (DL) models on a real-life dataset of obesity patients. The dataset comprised several patient health records with 17 different elements related to obesity, classifying obesity levels into seven types. The study evaluated the accuracy of various models before and after applying the Synthetic Minority Over-sampling Technique (SMOTE). Before SMOTE, the TabNet (T) and XG-Boost (XGB) classifiers achieved high accuracies of 96.6% and 96.2%, respectively, outperforming Random Forest (RF) (94.8%), Multi-Layer Perceptron (MLP) (94.5%), Bagging (B) (94.07%), Decision Tree (DT) (93.6%), Support Vector Machine (SVM) (82.5%), K-Nearest Neighbor (KNN) (75.9%), Stochastic Gradient Descent (SGD) (68.6%), AdaBoost (AB) (28.4%), Stacking (S) (16.8%), and G-Boost (GB) (95.5%). After applying SMOTE, GB and XGB showed improved accuracies of 99.3% and 99%, respectively, surpassing RF (97.4%), Bagging (96.28%), DT (96.9%), SVM (90.3%), KNN (85.7%), SGD (67.6%), AB (34.9%), and Stacking (12.3%). Comparatively, the existing methods showed accuracies with GB (97.2%), DT (96.7%), RF (94.8%), SVM (43.4%), and AB (33.1%), while the proposed models exhibited superior performance: GB (99.3%), DT (96.9%), RF (97.4%), SVM (90.2%), AB (34.9%), XGB (99%), TabNet (98.4%), and MLP (97.7%). The proposed models significantly outperformed the existing ones, demonstrating their effectiveness in predicting obesity levels. |
