Métodos estatísticos de proteção de dados condenciais sob a condição de Dierential Privacy

Abstract

Theamountofdataproducedindigitalerahasincreasedinthelastdecades. Awareof this, companies and organizations have been making all necessary eorts to analyze this amount of information. However, the attention concerning privacy of individuals records is increasing. In this sense, the data privacy area emerges with the goal to guarantee users anonymity in researches. Given that, this work shows anonymization methods for binary and categorical data, using the concept of dierential privacy synthetic data. We also present inferential techniques to analyze this kind of data. First, we recreate and complement the scenarios proposed by Charest (2011) to binary anonymized data. We then extend the model to categorical variables. Lastly, we apply the anonymization and inferential techniques to a real dataset of car insurance claims in Brazil in 2016 for the metropolian region of Belo Horizonte and Zona da Mata. On the results, we noticed that there is some information loss when the methodology of dierential privacy synthetic data is applied. However, using the appropriate techniques to make inference can provide accurate estimates.