Paper Details

The field of privacy has seen rapid advances inrecent years because of the increase in the ability to storedata. In particular, recent advances in the data mining fieldhave lead to increased concerns about privacy. While thetopic of privacy has been traditionally studied in the contextof cryptography and information hiding, recent emphasis ondata mining has lead to renewed interest in the field.A fruitful direction for future data mining research will be thedevelopment of techniques that incorporate privacy concerns.Specially, we address the following question. Since the primarytask in data mining is the development of models aboutaggregated data, can we develop accurate models without accessto precise information in individual data records? We considerthe concrete case of building a decision-tree classier fromtraining data in which the values of individual records have beenperturbed. The resulting data records look very different from theoriginal records and the distribution of data values is also verydifferent from the original distribution. While it is not possible toaccurately estimate original values in individual data records, wepropose a novel reconstruction procedure to accurately estimatethe distribution of original data values. By using thesereconstructed distributions, we are able to build classifiers whoseaccuracy is comparable to the accuracy of classifiers built withthe original data.

Anu Thomas, Jimesh Rana, “A Review on Privacy Preserving Data Mining Approaches”, International Journal of Advance Engineering and Research Development (IJAERD), Vol. 2, Issue 13, pp. -, January 2015.