Risk Prediction of Diabetic Readmission Based on Cost Sensitive Convolutional Neural Network

Diabetes is a chronic disease that nearly affects people of all ages. Some scholars find that the potential risk of diabetes can be effectively predicted by modeling the readmission data. However, a large number of data missing and data imbalance problems exist in diabetes medical data. Traditional machine learning approaches employ feature extraction and classification prediction. However, the huge problems caused by imbalances in medical data and the different costs of category labels and misclassification errors are not fully taken into account. In this paper, we propose a Cost Sensitive Convolutional Neural Network (CSCNN) model for the imbalanced diabetes dataset. We combined convolutional neural networks (CNN) and cost sensitive loss function to deal with data imbalance, and conducted sufficient experiments on the diabetes readmission dataset. Compared with other methods, our model has achieved good results in all aspects of indicators. In experiments, the F3 score of the model reaches 0.584 and the sensitivity reaches 0.782. We find that our proposed model can effectively classify the inpatient data of unbalanced diabetes and solve imbalance and skewness problems effectively.