Reinforcement Technique for Classifying Quasi and Non-quasi Attributes for Privacy Preservation and Data Protection

A quasi attribute refers to a distinct subset of unique attributes that can adequately recognize tuples in a table. Hasty distribution of the quasi attributes will prompt privacy leakage. Choosing private data from a list of attributes is decided by the publisher, and it undoubtedly changes from dataset to dataset. The need for dynamically choosing and informing systems about a quasi and a non-quasi attribute remains a challenging task. Presently, there is no particular automation model for the classification of quasi and non-quasi. It could be a burden when a massive dataset has to be classified, or aggregation of datasets has to be performed. This research paper considers the need to categorize quasi attributes for a non-expert through a direct attack and proposes a solution through the game theory approach and reinforcement machine learning model. For demonstration, a 2 × 2 state matrix is considered. The results include case-wise time consumption and comparison among all necessary steps for accurate navigation, between various attributes. Among all the notable cases, the matrix arrangement with a quasi attribute in 00 ^th and 11 ^th position, non-quasi in 01 ^th and 10 ^th position obtained better performance. This reinforcement-based solution helps the automation of the classification of quasi and non-quasi attributes.