Learning Decision Trees from Anonymized Data

DOI: 10.5176/2251-2136_ICT-BDCS17.33

Authors: Jianhua Shao and Jasmin Beckford.

Abstract:

There is much interest in developing solutions for protecting data privacy in recent years, and many privacy models and data sanitization methods have been proposed. However, relatively little has been done to understand how existing data analysis techniques may be adapted to work with sanitized data. In this paper we report a study on learning decision trees from anonymized data. We sanitize data using the Mondrian algorithm to satisfy k-anonymity and adapt the ID3 algorithm to learn decision trees from sanitized data. Our preliminary experiments show that accurate decision trees can be learnt from anonymized data, and degradation of classification accuracy is no more than 2{6e6090cdd558c53a8bc18225ef4499fead9160abd3419ad4f137e902b483c465} with typical settings.

Keywords: