Preserve the privacy of personal information when conducting statistical analysis has attracted much attention in machine learning and data mining. In this work, we propose an approach to realize privacy-preserving logistic regression when data are held by different individuals-without actually combining the data together. In our approach, we use polynomial fitting to approximate the logistic function in order to solve the problem that logistic function is not available in the secure settings (encryptions are not applicable) because of its non-linear property. And the experiment shows that our approach achieves good prediction accuracy compared with original logistic regression.
展开▼
