Automated disease diagnosis systems are essential to detect and classify diseases in a very short time. This study designed an automated model to classify diabetes into 3 classes: no diabetes, prediabetes, and diabetes. The model was constructed of self-organizing maps (SOM) with a classification model. SOM maps were used to map the data into a two-dimensional feature map. The SOM maps were performed for the hexagonal grid 3 \(\,\times \,\) 2 and a 3 \(\,\times \,\) 3 grid. The clustered samples in each neuron cell are used to build the following classification models: Decision Tree (DT), KNN algorithm, and Bagging decision tree (with 50, 100, and 150 subtrees). The classification accuracy was used to evaluate the efficiency of the tested models. The better classification results were achieved when combining a 3 \(\,\times \,\) 3 SOM map with bagging DT (150 trees) was about 84.81%. While the worst results were obtained from the KKN algorithm combined with 3 \(\,\times \,\) 2 SOM maps were about 77.53%. The execution time was measured in the test phase. The decision tree (DT) took the shortest time, while the KNN algorithm took the longest.