Predicting diabetes diseases using mixed data and supervised machine learning algorithms

Diabetes is considered as one of the deadliest and chronic diseases in several countries. All of them are working to prevent this disease at early stage by diagnosing and predicting the symptoms of diabetes using several methods. The motive of this study is to compare the performance of some Machine Learning algorithms, used to predict type 2 diabetes diseases. In this paper, we apply and evaluate four Machine Learning algorithms (Decision Tree, K-Nearest Neighbours, Artificial Neural Network and Deep Neural Network) to predict patients with or without type 2 diabetes mellitus. These techniques have been trained and tested on two diabetes databases: The first obtained from Frankfurt hospital (Germany), and the second is the well-known Pima Indian dataset. These datasets contain the same features composed of mixed data; risk factors and some clinical data. The performances of the experimented algorithms have been evaluated in both the cases i.e. dataset with noisy data (before pre-processing/some data with missing values) and dataset set without noisy data (after preprocessing). The results compared using different similarity metrics like Accuracy, Sensitivity, and Specificity gives best performance with respect to state of the art.