Prediction of diabetes using Multi-type data

The incidence rate of diabetes has increased dramatically. However, there are still some problems in diagnosing diabetes based on blood glucose and other Physical examination target. For example, trauma or other stress conditions can cause temporary hyperglycemia, which can disrupt the diagnosis of diabetes. In order to solve above problems, a large-scale word vector model in the field of diabetes is constructed to better represent the diabetes related words, and a new hybrid prediction model called Diabetes hybrid convolution neural network (DhCNN) is proposed. The model integrates numerical physical examination data and patient's description of their own symptoms, which makes up for the defect that only one type of data leads to insufficient features in diabetes data. More features also enable the convolution layer to capture more disease features, thus improving the accuracy of diabetes prediction. Experimental results on multiple datasets show that DhCNN has better performance in prediction accuracy than machine learning-based models and deep learning models that use a single type of data, especially in datasets with insufficient number of features.