OAN: aykırı kayıt yönelimli fayda temelli mahremiyet koruma modeli

Abstract

Developing a new privacy preserving model  Utility-based model  Outlier record-oriented model Figure A. General representation of the proposed model Purpose: In this paper, a new outlier record-oriented utility-based privacy preserving model was proposed. The existence of outliers in data set decreases data utility in anonymization. Hence, outliers should be managed in the anonymization process. In traditional management approaches, outliers are detected after anonymization and they are partially or completely removed from the published data set. However, detection of outliers after anonymization increases computational cost and the removal of the outliers from the data set reduces total data utility. In this study, a new outlier-oriented utility-based privacy preserving model named as OAN, which reduces the computational cost by detecting outliers before anonymization and increases data utility by using all data, was proposed. Theory and Methods: In the proposed model, data is divided into two subsets which are named as utility record set (URS) and outlier record set (ORS). URS includes normal data that presents high data utility. ORS is the set of outlier records that decreases total data utility. In order to increase total data utility, ORS is divided into two new subsets such as URS and ORS sets recursively. If a stopping criteria is met, this recursion stops and finally anonymized data is released. Local Outlier Factor was employed for outlier detection and anonymization was performed by using Mondrian algorithm. DM and AECS metrics were used to evaluate the information loss of the proposed model. Results: In the experiments, Adult data set was used to test the proposed model. DM and AECS metrics were employed to measure data utility. It was observed that the results of the proposed model presented better results than classical Mondrian. The results showed that OAN increases total data utility while preserving data privacy. In addition, it was showed that the proposed model is computational cost-effective compared to another utility based anonymization model. Conclusion: In this paper, a new outlier record-oriented utility-based privacy preserving model was proposed, tested and verified. Two information metrics and Adult data set were employed in the experiments and the results showed that the proposed model is an effective solution in terms of computational cost and data utility.

Derin Öğrenmede Diferansiyel Mahremiyet

“…Literatürde veri mahremiyetine yönelik çok sayıda tehdit, saldırı veya olumsuz durum mevcuttur. Bunların başında arka plan bilgileri ile veri bağlama (eşleştirme) yöntemleri kullanılarak yapılan veri ifşası [4,19,20] gelir. Arka plan bilgisi ile yayınlanan veriler arasında kayıt, hassas öznitelik veya tablo düzeyinde bağlantı kurarak çeşitli saldırılar düzenleyebilir [21,22].…”