Preserving Privacy by Quantizing

Advances in Data Mining resulted in collection of sensitive information from the published data. The web rendering a platform for data publishing and development of automated software technologies added fuel to the burning problem of personal privacy. The sensitive data need to be anonymized and published in the web. A number of methods were proposed earlier for anonymization. Methods include data partitioning, data swapping, generalization, suppression, randomization, perturbation, secure multiparty computation etc. A method of perturbation is discussed in this paper. Domain values of the private table are clustered together using clustering algorithms. To anonymize the private table the values are represented by the cluster head. This decreases the utility of data to be published. Care need to be taken that while anonymizing a balance of utility and privacy need to be maintained. F-measure and distortion are the metrics deployed to find the utility of the data that get perturbed.