Scalability and performance of the Privacy-Aware Classification method GenPAC

In this work we evaluate the scalability and performance of our previously presented GenPAC method by applying it on larger datasets. The work is motivated by the necessity of meeting privacy constraints when focusing on the importance and broad application of data mining but also by the growing demand for privacy preservation in general. GenPAC, which can be used with any standard classification method, relies on clustering data to obfuscate information. The method is particularly useful in multi-party data mining scenarios where privacy is of interest. Its application has only minor impact on the classification performance of the used underlying data mining method whilst privacy preservation can be provided at the cost of a higher execution time. In this regard, empirical analysis and evaluation have been conducted. The corresponding results are presented, analyzed and discussed with respect to their classification performance and execution time showing a high scalability in regard to the size of the dataset and the number of participating parties.