Decision Tree Evaluation on Sensitive Datasets for Secure E-Healthcare Systems
IEEE Transactions on Dependable and Secure Computing
Collecting and analyzing patients' e-healthcare data in Medical Internet-of-Things (MIOT), e-Healthcare providers can offer reliable medical services that will achieve better treatment for patients. For example, the diagnosis of disease and predictions of health offer an alternative and helpful evaluation of the risk of diseases, thereby helping patients lead a healthier life. However, e-Healthcare providers cannot cope with the huge volumes of data and respond to this online service such that a feasible solution is adopted to outsource the medical data to powerful medical cloud servers. Since medical data are very sensitive and outsourced servers are not fully trusted, a direct outsourcing decision tree evaluation service will inevitably result in huge privacy risks with regards to patient identity or original medical data. It is hard to hide the results of an evaluation from the single-server model unless a fully homomorphic cryptosystem is used, or the requester must communicate online with the cloud multiple times; and even then the efficiency issues between outsourced servers and patients must also be considered. With regards to these issues, this paper proposes a Secure and Privacy-Preserving Decision Tree Evaluation scheme (namely SPP-DTE) to achieve the secure disease diagnosis classification under e-Healthcare systems without revealing the sensitive information of patients such as physiological data or the private data of medical providers such as the structure of decision trees. Our proposed scheme uses modified KNN computation to match the similarity and preserve the confidentiality of raw data, and also applies matrix randomization and monotonically increasing and one-way function to confuse the intermediate results. The experiment is conducted in data sets from UCI machine learning repository of medical health data, our analysis indicates that the proposed SPP-DTE scheme is efficient in terms of computational cost and communication overhead that is practical and efficient for privacy protection in e-Healthcare classification and diagnosis system.
Open Access Status
This publication is not available as open access