Diabetes Prediction Using Machine Learning

Machine learning (ML) techniques for healthcare informatics provide health professional insight into disease development. Many healthcare topics are suitable for ML research, such as diabetes prediction and classification. Common ML approaches use a classification method to predict the outcome of the disease for given test data, though these solutions tend to have limited accuracy rates. Further tuning with extra manipulation of the dataset helps improve the model's accuracy to a certain level, but this requires certain professional knowledge in the medical domain. In this research, we propose using a DNN (Deep Neural Network) approach to predict the outcome of diabetes from the test data. Based on the dataset statistics, we simply transform 1D diabetes test data arrays to 2D Farrays without complex medical knowledge. We use a 2D convolution function to extract the features for prediction in addition to modifying the final stage activation function, to which the response is similar to a unit step function for binary classification problems. Our DNN model prediction accuracy has improved over the known non-deep learning classification models.