Prediction of daily patient visits in respiratory department using deep learning

Since climate changes and air pollution have a direct impact on the number of emergency respiratory patients visited, in this paper, we study predicting methods of daily patient visits in respiratory department based on deep learning. We validate our methods via experiments on the actual data of the First Affiliated Hospital of Xi'an Medical University. To improve the accuracy of prediction, we propose four data preprocessing methods, including meteorological lag and time sensitive method, sliding window method, seasonal classification method, as well as correlation analysis method. Experimental results show that the prediction accuracy of the deep learning model is improved from 76.28% to 86.37% after adopting our four preprocessing methods. The maximum error between predicted and actual visits is less than 20.