<![CDATA[Diabetes mellitus is a chronic disease with increasing prevalence and requires effective early detection efforts. This study aims to develop a diabetes risk prediction model using an Artificial Neural Network (ANN) based on non-laboratory health indicators. The dataset used is the CDC Diabetes Health Indicators with a large amount of data and characteristics of classes that are not fully balanced. The research stages include data preprocessing that includes handling missing values, encoding categorical data using one-hot encoding, normalization of numerical features, and analysis of the target class distribution. The ANN model was trained using a Multilayer Perceptron architecture with dropout regularization and L2 penalty and AdamW optimization. The evaluation results show that the model achieved an accuracy of 86.45%, a precision of 85.2%, a recall of 82.7%, and an AUC-ROC value of 0.89. Although the accuracy is in the medium range for a large dataset, the high AUC value indicates excellent model discrimination ability. This performance is affected by the limited number of non-laboratory features used and the imbalanced class distribution. The findings of this study indicate that ANN based on simple health indicators has the potential to be used as a diabetes risk screening tool in primary healthcare. Further research is recommended to apply class balancing techniques, model interpretability analysis, and external validation in the Indonesian population.]]>
