Privacy-Preserving Self-Organizing Map

Privacy-preserving data mining seeks to allow the cooperative execution of data mining algorithms while preserving the data privacy of each party concerned. In recent years, many data mining algorithms have been enhanced with privacy-preserving feature: decision tree induction, frequent itemset counting, association analysis, k-means clustering, support vector machine, Näıve Bayes classifier, Bayesian networks, and so on. In this paper, we propose a protocol for privacypreserving self-organizing map for vertically partitioned data involving two parties. Self-organizing map (SOM) is a widely used algorithm for transforming data sets to a lower dimensional space to facilitate visualization. The challenges in preserving data privacy in SOM are (1) to securely discover the winner neuron from data privately held by two parties; (2) to securely update weight vectors of neurons; and (3) to securely determine the termination status of SOM. We propose protocols to address the above challenges. We prove that these protocols are correct and privacy-preserving. Also, we prove that the intermediate results generated by these protocols do not violate the data privacy of the participating parties.