QIDLearningLib: A Python library for quasi-identifier recognition and evaluation

Quasi-identifiers (QIDs) are attributes in a dataset that are not directly unique identifiers of the users/entities themselves but can be used, often in conjunction with other datasets or information, to identify individuals and thus present a privacy risk in data sharing and analysis. Identifying QIDs is important in developing proper strategies for anonymization and data sanitization. This paper proposes QIDLearningLib , a Python library that offers a set of metrics and tools to measure the qualities of QIDs and identify them in data sets. It incorporates metrics from different domains – causality, privacy, data utility, and performance – to offer a holistic assessment of the properties of attributes of a given tabular dataset. Furthermore, QIDLearningLib offers visual analysis tools to present how these metrics shift over a dataset and implements an extensible framework that employs multiple optimization algorithms such as anevolutionary algorithm, simulated annealing, and greedy search using these metrics to identify a meaningful set of QIDs. • QIDLearningLib is the first library for automated QID recognition in tabular datasets. • It integrates metrics from causality, data privacy, and data utility for flexible QID selection. • It provides metrics to evaluate the performance of the QID selection system, against a ground-truth. • Includes multiple optimization algorithms for QID selection based on user-defined metrics. • Supports redundancy analysis to identify the most relevant, non-overlapping metrics. • Provides graphical and testing tools for enhanced interpretability.