Health Risk Prediction Using Big Medical Data - a Collaborative Filtering-Enhanced Deep Learning Approach
Xin Li, Juan Li · 2018
The massive amount of medical data accumulated from patients and healthcare providers has become a vast reservoir of knowledge source that may enable promising applications such as risk predictive modeling, clinical decision support, disease or safety surveillance. However, discovering knowledge from the big medical data can be very complex because of the nature of this type of data: they normally contain large amount of unstructured data; they may have lots of missing values; they can be highly complex and heterogeneous. To address these challenges, in this paper we propose a Collaborative Filtering-Enhanced Deep Learning approach. In particular, we estimate missing values based on patients' similarity, i.e., we predict one patient's missing features based on the values of similar patients. This is implemented with the Collaborative Topic Regression method, which tightly couples topic model and probability matrix factorization and is able to utilize the rich information hidden in the data. Then a deep neural network-based method is applied for the prediction of health risks. This method can help us handle complex and multi-modality data. Extensive experiments on a real-world dataset have been performed and the results show improvements of our proposed algorithm over the state-of-the-art methods.