PPDP: An efficient and privacy-preserving disease prediction scheme in cloud-based e-Healthcare system

Disease prediction systems have played an important role in people's life, since predicting the risk of diseases is essential for people to lead a healthy life. The recent proliferation of data mining techniques has given rise to disease prediction systems. Specifically, with the vast amount of medical data generated every day, Single-Layer Perceptron can be utilized to obtain valuable information to construct a disease prediction system. Although the disease prediction system is quite promising, many challenges may limit it in practical use, including information security and prediction efficiency. In this paper, we propose an efficient and privacy-preserving disease prediction system, called PPDP. In PPDP, patients' historical medical data are encrypted and outsourced to the cloud server, which can be further utilized to train prediction models by using Single-Layer Perceptron learning algorithm in a privacy-preserving way. The risk of diseases for new coming medical data can be computed based on the prediction models. In particular, PPDP builds on new medical data encryption, disease learning and disease prediction algorithms that novelly utilize random matrices. Security analysis indicates that PPDP offers a required level of privacy protection. In addition, real experiments on different datasets show that computation costs of data encryption, disease learning and disease prediction are several magnitudes lower than existing disease prediction schemes.