Obesity has become a dominant health concern globally and it is surpassing undernutrition and contagious diseases as a major contributor to ill health issues. This alarming trend highlights the need for effective prediction and prevention strategies. Addressing obesity is required to create supportive environments that encourage healthy living. It is known that multiple complex factors influence the onset of obesity, thus presenting a challenge for the prediction and diagnosis of obesity. The current study uses several machine-learning (ML) techniques for predicting obesity levels. The dataset used contains data for the estimation of obesity levels in individuals from three Latin American countries which are Mexico, Peru, and Colombia, based on their dietary patterns and physical activity levels. It comprises 2111 records, with 17 attributes, each attribute representing an individual factor that could contribute to obesity. These predictor attributes include demographic information like age and gender, various lifestyle choices. The records were labeled in various classes, i.e. insufficient weight, normal weight, overweight-I and II, obesity type-I, II, and III. Various ML methods like decision trees, Support Vector Machine (SVM), Ensemble models, artificial neural networks (ANN), and kernels were utilized. The performance metrics of all the models are compared. The findings indicate that the ANN and SVM techniques show the best validation accuracy of 96% in predicting obesity levels.