Global Knowledge, Local Impact: Domain Adaptation and Classification for Obesity in the UAE

Obesity, a global public health concern, is escalating rapidly, especially in the Middle East, with the United Arab Emirates (UAE) witnessing one of the highest prevalence rates among adults and children. This multifactorial health issue is influenced by genetic, environmental, behavioral, and social factors. However, the challenge lies in the lack of comprehensive and representative data that encapsulates the diversity and complexity of the population, particularly in the UAE. This research paper presents a novel AI-driven approach to address the obesity problem in the UAE. The study employs state-of-the-art data augmentation techniques to generate local data that are realistic, diverse, and representative of the UAE population and the global obesity situation. The data is derived from four global datasets, two related to obesity and two related to diabetes. The paper also identifies key features and factors that influence obesity in the UAE using machine learning and feature extraction methods. A predictive model is built and evaluated using the local data and the best-performing classifier, the random forest classifier. Our study’s random forest classifier achieved a 94.83% accuracy when trained only on the global dataset. However, when fine-tuned on datasets synthesized with different augmentation methods, it showed better results: SMOTE excelled with 98.09% accuracy, TabGAN was close behind at 97.02%, and VAE lagged at 70.99%. Therefore, domain adaptation using the synthesized data confirmed the robustness of the generated data, with SMOTE and TabGAN outperforming the original dataset with accuracies of 98.09% and 97.22%, respectively.