Predicting medical codes via deep learning-based methods

Deep learning methods are used for prediction tasks in many areas such as computer vision, speech recognition, and natural language processing. Such methods give more flexible and powerful results than traditional machine learning methods. Improvements in the publicly available healthcare data sets lead to deep learning works to be increased in the healthcare domain especially for predicting the medical codes of patients. In this work, a deep learning-based multi-modal method is proposed to predict medical codes by using both text-based and time-series data. This method was developed with the help of publicly available MIMIC-III data set and gives competitive results with similar healthcare prediction methods which are using only text-based data. Furthermore, the proposed approach is also tested with some state-of-the-art text-based models and showed that using both types of data improves results in terms of F1 score.