Fuzzy-based Methods for Privacy-Preserving Data Mining

As more and more organizations are collecting and sharing data about their customers, there is a growing concern about violation of customer privacy. While some of the sharing is for the benefit of general public such as to understand disease behavior in medical research, individuals are concerned about violation of their privacy. The middle ground is found through privacy-preserving data mapping. Here, sensitive attributes of data are mapped to another domain so that original values are not revealed and yet the original associations are retained. In this paper, we compare a set of fuzzy-based mapping techniques in terms of their privacy-preserving property and their ability to retain the same relationship with other fields. In particular, our contribution is on four fronts: (i) modification of the fuzzy function definition, (ii) introducing seven ways to combine the different functional values for a data item into a single value, (iii) using several similarity metrics to compare the original data with the mapped data, (iv) measuring the effect of mapping on derived association rule. The paper presents preliminary results in this direction and proposes future work in this area.