Background: Diabetes mellitus (DM) is a chronic metabolic disorder with an increasing global prevalence that requires early detection and effective predictive models. The use of artificial intelligence in disease diagnosis and management has significantly increased in recent years. Artificial neural network (ANN) models playing a crucial role in enhancing predictive accuracy and clinical decision-making. This study compares the performance of Radial Basis Function (RBF) networks, Recurrent Neural Networks (RNNs), and Extreme Learning Machines (ELMs) for diabetes prediction. Methods: A case-control study was conducted on 800 participants (400 diabetics and 400 non-diabetics) recruited from a hospital outpatient department. The data were collected on demographic, lifestyle, anthropometric, and medical history variables using structured case record sheet.Predictive models were developed in RStudio and trained on 80% of the dataset, with 20% reserved for testing. Model performance was assessed using accuracy, sensitivity, specificity, Kappa score, and Area Under the Receiver Operating Characteristic (AUROC) curve. Results:In the prediction of diabetes, ELM outperformed RBF and RNN models in terms of accuracy (80.00%), sensitivity (77.50%), specificity (82.50%), Kappa score (0.60), and precision (81.58 percent).RBF attained moderate accuracy (74.38%), whereas RNN showed lower predictive performance (70.62%), likely due to its suitability for sequential data rather than structured medical datasets. Conclusion: ELM demonstrated superior predictive capability over RBF and RNN in diabetes classification, making it the recommended model for risk assessment.