Abstract:

Diabetes is a metabolic disease that occurs due to high blood sugar levels in the body. If it is not treated, diabetes-related health problems may occur in many vital organs of the body. With the latest techniques in machine learning technologies, some of the applications can be used to diagnose diabetes at an early stage. In this study, the data set from the laboratories of Medical City Hospital Endocrinology and Diabetes Specialization Center Al Kindy Training Hospital was used. The dataset consists of 3 different classes: normal, pre-diabetes and diabetes. The obtained diabetes dataset was classified using Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) deep learning methods. The classification performance of each algorithm was evaluated with accuracy, precision, sensitivity and F score performance parameters. Among the deep learning methods, 96.5% classification accuracy was obtained with the LSTM algorithm, 94% with the CNN algorithm and 93% with the GRU algorithm. In this study, the Permutation Feature Importance (PFI) method was also used to determine the effect of features in the data set on classification performance. With this method, study reveals that the HbA1c feature is an important parameter in the used deep learning methods. Both the results obtained with the LSTM algorithm and the determination of the most important feature affecting the classification success reveal the originality of the study. It shows that the obtained results will provide healthcare professionals with a prognostic tool for effective decision-making that can assist in the early detection of the disease.

Keywords: