Childhood obesity is a major public health challenge worldwide, with increasing prevalence and significant long-term health consequences. This study aims to classify obesity phenotypes using a machine learning-based clustering approach and assess the role of anthropometric, inflammatory and metabolic risk factors in identifying cardiovascular high-risk individuals. A cohort of 1,461 school children and adolescents was analysed using the biomarkers for these factors. Clustering techniques, including Principal Component Analysis and K-means, were applied to identify distinct obesity subtypes. A stacking ensemble model incorporating Logistic Regression, Random Forest, CatBoost, and Gradient Boosting classifiers was used for predictive analysis. Feature importance was assessed to determine key contributors to obesity risk stratification. Three distinct obesity subtypes were identified: metabolically healthy obese, metabolically unhealthy obese, and normal-weight individuals. Metabolically unhealthy obese individuals exhibited significantly higher levels of C-reactive protein, central adiposity, and metabolic abnormalities, confirming the role of systemic inflammation in obesity-related cardiometabolic dysfunction. The predictive model achieved an area under the receiver operating characteristic curve of 0.93, demonstrating strong classification accuracy. Feature importance analysis revealed that fat mass, muscle mass, inflammatory markers, and parental health history were key predictors of obesity phenotypes. This study highlights the heterogeneity of childhood obesity and underscores the need for a multidimensional classification framework beyond BMI alone. The integration of machine learning methodologies enhances obesity risk stratification, enabling early identification of high-risk individuals.