Classification of Adults' Obesity and Overweight using Machine Learning Models: Case Study Palestine رسالة ماجستير

Abstract

Obesity and Overweight are considered a major cause of many illnesses on the global level. According to the World Health Organization (WHO), more than 30% of global population are suffering Obesity and Overweight. This percentage may increase in the coming years. This global phenomenon would lead to a series of dangers considering the emergence of other diseases. Obesity is often associated with other chronic diseases, such as arterial hypertension, and type 2 diabetes mellitus. The risk of suffering from these comorbidities is greater as the body mass index increases. However, there are infrequent studies investigating obesity in Palestine. Such lack of knowledge affects the prediction identification and prevention of obesity in Palestinian society. As a result, this research aims to fill this gap in obesity research in Palestine, depending on real risk factors related to Palestinian society. The research dataset was collected from Palestine which included 902 participants. However, after classifying the participants into four categories, Underweight, Normal, Overweight, and Obesity, the first class was imbalanced which required using data balancing methods in this work we use Synthetic Minority Over-sampling Technique (SMOTE). The collected data was processed by ML models. The models include support vector machine (SVM), Random Forest (RF), decision tree (DT) multi-layer perceptron (MLPNNs), XGBoost, Adaboost, Extra tree classifier, and Gradient Boost (GB). The grid search was applied to models to obtain the appropriate set of parameters for each model. The performance of the models was evaluated on imbalance and balanced data, and the model that outperforms other models with the highest accuracy is the RF with (98.3%, and 98.6%) accuracy respectively