Privacy-aware dynamic feature selection

Abstract

Big data will enable the development of novel services that enhance a company's market advantage, competition, or productivity. At the same time, the utilization of such a service could disclose sensitive data in the process, which raises significant privacy concerns. To protect individuals, various policies, such as the Code of Fair Information Practices, as well as recent laws require organizations to capture only the minimal amount of data necessary to support a service. While this is a notable goal, choosing the minimal data is a non-trivial process, especially while considering privacy and utility constraints.In this paper, we introduce a technique to minimize sensitive data disclosure by focusing on privacy-aware feature selection. During model deployment, the service provider requests only a subset of the available features from the client, such that it can produce results with maximal confidence, while minimizing its ability to violate a client's privacy. We propose an iterative approach, where the server requests information one feature at a time until the client-specified privacy budget is exhausted. The overall process is dynamic, such that the feature selected at each step depends on the previously selected features and their corresponding values. We demonstrate our technique with three popular classification algorithms and perform an empirical analysis over three real world datasets to illustrate that, in almost all cases, classifiers that select features using our strategy have the same error-rate as state-of-the art static feature selection methods that fail to preserve privacy.

PUTWorkbench: Analysing Privacy in AI-intensive Systems

“…(put number = -0. Third, there are relatively recent works, which are precursor to our work, that propose the idea of privacy-aware feature selection [26][27] [7][39] [38] [34]. Our work contrasts these works in either one or both of the following ways.…”

Test time feature ordering with FOCUS

“…Finally, there are methods that determine a feature order at test time, using the expected quality of the subsequent prediction to decide which feature to acquire next. For example, Pattuk et al [33] formulate a privacy-aware dynamic feature selection algorithm for classification that sequentially chooses features for a test instance, according to which will most increase the expected confidence of the next prediction, as long as including that feature does not violate a privacy constraint. This work is most responsive to the test time situation.…”

“…The Utility Function used to determine the value of a feature also varies. Examples for feature "quality" include subsequent prediction accuracy (e.g., [21,41,49,51]) and subsequent prediction uncertainty (e.g., [33]). Finally, the Domain column shows where each relevant approach was applied.…”

“…For application areas where feature computation time is a bottleneck at test time (e.g., natural language processing (NLP), computer vision), the primary goal of test time feature acquisition is to speed up prediction (e.g., [21,42,49]). In other domains, there is an allowable test time budget of costs other than time, such as user burden of providing information (e.g., [12]) or the resulting loss in privacy from disclosing information (e.g., [33]).…”

Feature Selection for Classification under Anonymity Constraint

“…For the numerical attributes, we cut them into different categories and consider a binary attribute for each category. For instance, we partition age value in five non-overlapping intervals: [0, 25], (25,35], (35,45], (45,55], and (55, ∞], and then each of the five intervals becomes a binary attribute. Similarly, education attribute is divided into 4 intervals and hour/week attribute is divided into 5 interval.…”