Abstract:We develop a model using deep learning techniques and natural language processing on unstructured text from medical records to predict hospital-wide $30$-day unplanned readmission, with c-statistic $.70$. Our model is constructed to allow physicians to interpret the significant features for prediction.