This study implements a multilayer perceptron (MLP) neural network for the early prediction of type 2 diabetes mellitus (T2DM), including patients with normal fasting glucose levels, to optimize medical decision-making based on clinical data and established guidelines. Using data from a hospital information system (HIS) encompassing 40,246 samples, the MLP neural network (MLPClassifier) was trained through fivefold cross-validation. Key performance metrics, including accuracy, sensitivity, and specificity, were evaluated for diabetic (G2) and non-diabetic (G1) patients. The model achieved moderate performance, with an average accuracy of 0.78 and an F1-score of 0.67. Accuracy for non-diabetic patients (G1) was 88%, whereas for diabetic patients (G2), it was 30%, with an overall sensitivity of 68%. Despite the class imbalance, the neural network managed the classification effectively, achieving an area under the ROC curve (AUC-ROC) of 0.71. These results position the MLP model as a promising tool for early T2DM detection and medical decision support. Future efforts to enhance model precision for diabetic patients should focus on hyperparameter optimization and sampling techniques, emphasizing its potential for application in resource-limited healthcare settings.