Mathematics & Physics 2015, 8(4), 478–486 УДК 004.056.5 Privacy-Preserving Building of Self-Organizing Maps Alexey V.Vashkevich∗ Vadim G.Zhukov† Eugene S. Semenkin‡ Institute of Mathematics University of Potsdam Am Neuen Palais, 10, Potsdam, 14469 Germany Received 03.06.2015, received in revised form 09.07.2015, accepted 24.08.2015 Various data mining techniques are designed for extracting significant and valuable patterns from huge databases. <...> Today databases are often divided between several organizations for the reason of limitations like geographical remoteness, but the most important limit is preserving privacy, unwillingness of data disclosing. <...> Every party involved in analysis wants to keep its own information private because of legal regulations and reasons of know-how. <...> Secure multiparty computations are designed for data mining execution in a multiparty environment, where it is extremely important to maintain the privacy of the input (and possibly output) data. <...> A self-organizing map is the data mining method by which analytics can display patterns on two-dimensional intuitive maps and recognize data clusters. <...> The protocols allow the implementation of a self-organizing map algorithm for two parties with horizontally partitioned data and for several parties with vertically partitioned data. <...> Keywords: secure multiparty computations, secure dot product, cluster analysis, self-organizing map. <...> In some cases of this research information may be split between several organizations or individual users. <...> This information may be private, so traditional conducting of a cluster analysis will disclose private data to all (or almost all) other participants. <...> The database partition may be horizontal, when each party owns a full dataset of some objects, or vertical, when each party owns some attributes of each object. <...> In these situations it is necessary to use secure multiparty protocols for processing private input data. <...> Building of self-organizing maps is one of the most common and used methods of data analysis in which analytics may face the problem of keeping the privacy of input data. <...> The main advantage of this data mining algorithm is the projection of multidimensional space on several two-dimensional maps, which show the evident final result. <...> This clustering algorithm is particularly relevant at the stage ∗alex23-5@yandex <...>