PENDA: Privacy ENhanced Data Aggregator

Abstract

Data is spread across different organizations and must be combined to get valuable analytics or train machine learning models. Sensitive features such as identification numbers, as columns, are common in these data, and organizations can only link if they share these columns. However, data protection regulations prevent these organizations from revealing the values in these columns to others. This paper proposes a technique, PENDA (Privacy ENhanced Data Aggregator), to encrypt columns of a database table or spreadsheet so that a central aggregation server can join them without decrypting using XOR-based homomorphism. We implement our PENDA system and demonstrate how organizations taking part in the process can encrypt and merge data. The experimental results show that the system can handle very large data files and scale to multiple organizations.