Model Prediction Using Artificial Neural Network (ANN) to Strengthen Diagnostic Analysis of Diabetes Melitus

The incidence of Diabetes Mellitus (DM) is one of the urgent and increasing health issues every year. Hence, this condition requires high urgency to be handled. The research aimed to develop a prediction model for DM that could be used in general for the purpose of diagnostic analysis of DM cases against suspected individuals. The dataset was sourced from the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK), which had closely related parameters in diagnostic analysis without favoring certain groups. The targeted contribution was the result of a new prediction model that was specifically tested on the dataset using the Artificial Neural Network (ANN) algorithm. This model was developed through a baseline model that was tested and improved in performance through hyperparameter cross-validation therapy and L1 regularization. The formation of the model architecture through experiments to adjust the conditions of hidden layers and neurons in several configurations results in a model architecture with 8 input parameters. It contains 3 hidden layers with a total of 14, 20, and 26 neurons, with the ReLU activation function on each hidden layer and the Sigmoid activation function on the output part. The second test is carried out on a hyperparameter configuration. It produces maximum performance with a k-fold value of 10 and L1 regularization of 0.0001. The model performance results obtain an accuracy value of 0.947, precision of 0.895, recall of 0.914, and model loss of 0.215