A privacy preserving neural network learning algorithm for horizontally partitioned databases

Ordinary data mining requires accurate input data, but privacy concerns may bar use of such techniques. Thus, privacy preserving data mining methods are needed, which can work well without opening the private data. Although, much work has been done on privacy preserving classification, to the best of our knowledge, there has not been a privacy preserving perceptron neural network learning algorithm that can work in the real world on distributed databases. To solve this problem, this study brings forward a privacy preserving Back Propagation (BP) learning algorithm for horizontally partitioned databases. In this algorithm, data nodes can privately exchange information that the original BP algorithm needs. This algorithm can obtain the same result as learning on global data using the BP algorithm without considering privacy protection and each data nodes is prevented from obtaining detailed data on other nodes in the learning process.