Disease Risk Prediction by Mining Personalized Health Trend Patterns: A Case Study on Diabetes

Health examination has played an important role for maintaining people's health since it can not only help people understand their own health conditions clearly but also avoid missing the best timing of disease treatment. However, in current health examination systems, people get only a basic report from single health examination and no advanced health risk analysis is provided. In this paper, we proposed an effective mechanism for chronic disease risk prediction by mining the data containing historical health records and personal life style information. Value change trends of the data are important for disease status prediction, and we defined significant ones as health risk patterns in our mechanism. Risks of a chronic disease can be predicted early with a mechanism built with our health risk patterns and it also proven work well through experimental evaluations on real datasets. Our method outperformed traditional mechanism in terms of accuracy, precision and sensitivity for predicting the risk of diabetes. In particular, insightful observations show that the consideration of life-style information can effectively enhance whole performance for risk prediction. Moreover, classification rules produced by our mechanism which integrates C4.5 and CBA provide physicians disease related health risk patterns such that appropriate treatments could be given to people for disease prevention.