A fruitful direction for future data mining research will be the developmen
t of techniques that incorporate privacy concerns. Specifically, we address
the following question. Since the primary task in data mining is the devel
opment of models about aggregated data, can we develop accurate models with
out access to precise information in individual data records? We consider t
he concrete case of building a decision-tree classifier from training data
in which the values of individual records have been perturbed. The resultin
g data records look very different from the original records and the distri
bution of data values is also very different from the original distribution
. While it is not possible to accurately estimate original values in indivi
dual data records, we propose a novel reconstruction procedure to accuratel
y estimate the distribution of original data values. By using these reconst
ructed distributions, we are able to build classifiers whose accuracy is co
mparable to the accuracy of classifiers built with the original data.