Provided by: Springer Healthcare
Date Added: Jul 2007
In this paper, the authors investigate the accuracy (in terms of bias and variance of estimates) of estimates of various rules derived from the randomized market basket data and present a general framework which can conduct theoretical analysis on how the randomization process affects the accuracy of various measures adopted in market basket data analysis. They also show several measures (e.g., correlation) have monotonic property, i.e., the values calculated directly from the randomized data are always less or equal than those original ones. Hence, some market basket data analysis tasks can be executed on the randomized data directly without the release of distortion probabilities, which can better protect data privacy.