Prediksi Diabetes Menggunakan Deep Neural Network dengan Penyesuaian Hiperparameter Berbasis Bayesian Optimization

Diabetes mellitus, one of the chronic diseaseswhose population continues to increase worldwide, is a major problemfor the health sector. It is veryimportant to detect this disease early to prevent more severe complications. By using Deep Neural Network (DNN) and Bayesian optimization-based hyperparameter adjustmentOptimization, this study aims to create an optimal diabetes prediction modelto improve the accuracy and stability of prediction results. The dataset used consists of 15,000patients classified into two categories: non-diabetes and diabetes. Todetermine the best combination of hyperparameters, such as training level, number ofneurons, and activation function, the Bayesian optimization method was used. The evaluation results show that the model has high performance with an Area Under Curve (AUC) value of 0.9764, an accuracy of 97.14%, a precision of 98.40%, a recall of 91.76%, and an F1 score of 0.8004 at anoptimal threshold of 0.894. In addition, the classification report shows that the modelcan detect the non-diabetes class well with a recall of 99.90%, andperforms well for the diabetes class as well. The results of this studyshow that the use of Deep Neural Networks together withBayesian Optimization-based hyperparameter tuning improves diabetes prediction results. This method can be developedinto an AI-based medical decision support system that can help health workersmake early diagnoses properly.