Faisal Kamiran

Discrimination Aware Decision Tree Learning

Recently, the following discrimination aware classification problem was introduced: given a labeled dataset and an attribute B, find a classifier with high predictive accuracy that at the same time does not discriminate on the basis of the given attribute B. This problem is motivated by the fact that often available historic data is biased due to discrimination, e.g., when B denotes ethnicity. Using the standard learners on this data may lead to wrongfully biased classifiers, even if the attribute B is removed from training data. Existing solutions for this problem consist in “cleaning away” the discrimination from the dataset before a classifier is learned. In this paper we study an alternative approach in which the non-discrimination constraint is pushed deeply into a decision tree learner by changing its splitting criterion and pruning strategy. Experimental evaluation shows that the proposed approach advances the state-of-the-art in the sense that the learned decision trees have a lower discrimination than models provided by previous methods, with little loss in accuracy.

Classifying without discriminating

Classification models usually make predictions on the basis of training data. If the training data is biased towards certain groups or classes of objects, e.g., there is racial discrimination towards black people, the learned model will also show discriminatory behavior towards that particular community. This partial attitude of the learned model may lead to biased outcomes when labeling future unlabeled data objects. Often, however, impartial classification results are desired or even required by law for future data objects in spite of having biased training data. In this paper, we tackle this problem by introducing a new classification scheme for learning unbiased models on biased training data. Our method is based on massaging the dataset by making the least intrusive modifications which lead to an unbiased dataset. On this modified dataset we then learn a non-discriminating classifier. The proposed method has been implemented and experimental results on a credit approval dataset show promising results: in all experiments our method is able to reduce the prejudicial behavior for future classification significantly without loosing too much predictive accuracy.

Data preprocessing techniques for classification without discrimination

Recently, the following Discrimination-Aware Classification Problem was introduced: Suppose we are given training data that exhibit unlawful discrimination; e.g., toward sensitive attributes such as gender or ethnicity. The task is to learn a classifier that optimizes accuracy, but does not have this discrimination in its predictions on test data. This problem is relevant in many settings, such as when the data are generated by a biased decision process or when the sensitive attribute serves as a proxy for unobserved features. In this paper, we concentrate on the case with only one binary sensitive attribute and a two-class classification problem. We first study the theoretically optimal trade-off between accuracy and non-discrimination for pure classifiers. Then, we look at algorithmic solutions that preprocess the data to remove discrimination before a classifier is learned. We survey and extend our existing data preprocessing techniques, being suppression of the sensitive attribute, massaging the dataset by changing class labels, and reweighing or resampling the data to remove discrimination without relabeling instances. These preprocessing techniques have been implemented in a modified version of Weka and we present the results of experiments on real-life data.

Building Classifiers with Independency Constraints

In this paper we study the problem of classifier learning where the input data contains unjustified dependencies between some data attributes and the class label. Such cases arise for example when the training data is collected from different sources with different labeling criteria or when the data is generated by a biased decision process. When a classifier is trained directly on such data, these undesirable dependencies will carry over to the classifier’s predictions. In order to tackle this problem, we study the classification with independency constraints problem: find an accurate model for which the predictions are independent from a given binary attribute. We propose two solutions for this problem and present an empirical validation.

Decision Theory for Discrimination-Aware Classification

Social discrimination (e.g., against females) arising from data mining techniques is a growing concern worldwide. In recent years, several methods have been proposed for making classifiers learned over discriminatory data discrimination-aware. However, these methods suffer from two major shortcomings: (1) They require either modifying the discriminatory data or tweaking a specific classification algorithm and (2) They are not flexible w.r.t. discrimination control and multiple sensitive attribute handling. In this paper, we present two solutions for discrimination-aware classification that neither require data modification nor classifier tweaking. Our first and second solutions exploit, respectively, the reject option of probabilistic classifier(s) and the disagreement region of general classifier ensembles to reduce discrimination. We relate both solutions with decision theory for better understanding of the process. Our experiments using real-world datasets demonstrate that our solutions outperform existing state-of-the-art methods, especially at low discrimination which is a significant advantage. The superior performance coupled with flexible control over discrimination and easy applicability to multiple sensitive attributes makes our solutions an important step forward in practical discrimination-aware classification.

Quantifying explainable discrimination and removing illegal discrimination in automated decision making

Recently, the following discrimination-aware classification problem was introduced. Historical data used for supervised learning may contain discrimination, for instance, with respect to gender. The question addressed by discrimination-aware techniques is, given sensitive attribute, how to train discrimination-free classifiers on such historical data that are discriminative, with respect to the given sensitive attribute. Existing techniques that deal with this problem aim at removing all discrimination and do not take into account that part of the discrimination may be explainable by other attributes. For example, in a job application, the education level of a job candidate could be such an explainable attribute. If the data contain many highly educated male candidates and only few highly educated women, a difference in acceptance rates between woman and man does not necessarily reflect gender discrimination, as it could be explained by the different levels of education. Even though selecting on education level would result in more males being accepted, a difference with respect to such a criterion would not be considered to be undesirable, nor illegal. Current state-of-the-art techniques, however, do not take such gender-neutral explanations into account and tend to overreact and actually start reverse discriminating, as we will show in this paper. Therefore, we introduce and analyze the refined notion of conditional non-discrimination in classifier design. We show that some of the differences in decisions across the sensitive groups can be explainable and are hence tolerable. Therefore, we develop methodology for quantifying the explainable discrimination and algorithmic techniques for removing the illegal discrimination when one or more attributes are considered as explanatory. Experimental evaluation on synthetic and real-world classification datasets demonstrates that the new techniques are superior to the old ones in this new context, as they succeed in removing almost exclusively the undesirable discrimination, while leaving the explainable differences unchanged, allowing for differences in decisions as long as they are explainable.

Controlling Attribute Effect in Linear Regression

In data mining we often have to learn from biased data, because, for instance, data comes from different batches or there was a gender or racial bias in the collection of social data. In some applications it may be necessary to explicitly control this bias in the models we learn from the data. This paper is the first to study learning linear regression models under constraints that control the biasing effect of a given attribute such as gender or batch number. We show how propensity modeling can be used for factoring out the part of the bias that can be justified by externally provided explanatory attributes. Then we analytically derive linear models that minimize squared error while controlling the bias by imposing constraints on the mean outcome or residuals of the models. Experiments with discrimination-aware crime prediction and batch effect normalization tasks show that the proposed techniques are successful in controlling attribute effects in linear regression models.

Anti-discrimination analysis using privacy attack strategies

Social discrimination discovery from data is an important task to identify illegal and unethical discriminatory patterns towards protected-by-law groups, e.g., ethnic minorities. We deploy privacy attack strategies as tools for discrimination discovery under hard assumptions which have rarely tackled in the literature: indirect discrimination discovery, privacy-aware discrimination discovery, and discrimination data recovery. The intuition comes from the intriguing parallel between the role of the anti-discrimination authority in the three scenarios above and the role of an attacker in private data publishing. We design strategies and algorithms inspired/based on Frechet bounds attacks, attribute inference attacks, and minimality attacks to the purpose of unveiling hidden discriminatory practices. Experimental results show that they can be effective tools in the hands of anti-discrimination authorities.

Techniques for Discrimination-Free Predictive Models

In this chapter, we give an overview of the techniques developed ourselves for constructing discrimination-free classifiers. In discrimination-free classification the goal is to learn a predictive model that classifies future data objects as accurately as possible, yet the predicted labels should be uncorrelated to a given sensitive attribute. For example, the task could be to learn a gender-neutral model that predicts whether a potential client of a bank has a high income or not. The techniques we developed for discrimination-aware classification can be divided into three categories: (1) removing the discrimination directly from the historical dataset before an off-the-shelf classification technique is applied; (2) changing the learning procedures themselves by restricting the search space to non-discriminatory models; and (3) adjusting the discriminatory models, learnt by off-the-shelf classifiers on discriminatory historical data, in a post-processing phase. Experiments show that even with such a strong constraint as discrimination-freeness, still very accurate models can be learnt. In particular,we study a case of income prediction,where the available historical data exhibits a wage gap between the genders. Due to legal restrictions, however, our predictions should be gender-neutral. The discrimination-aware techniques succeed in significantly reducing gender discrimination without impairing too much the accuracy.

Bias-aware lexicon-based sentiment analysis

Sentiment analysis of textual content is widely used for automatic summarization of opinions and sentiments expressed by people. With the growing popularity of social media and user-generated content, efficient and effective sentiment analysis is critical to businesses and governments. Lexicon-based methods provide efficiency through their manually developed affective word lists and valence values. However, the predictions of such methods can be biased towards positive or negative polarity thus distorting the analysis. In this paper, we propose Bias-Aware Thresholding (BAT), an approach that can be combined with any lexicon-based method to make it bias-aware. BAT is motivated from cost-sensitive learning where the prediction threshold is changed to reduce prediction error bias. We formally define bias in polarity predictions and present a measure for quantifying it. We evaluate BAT in combination with AFINN and SentiStrength -- two popular lexicon-based methods -- on seven real-world datasets. The results show that bias reduces smoothly with an increase in the absolute value of the threshold, and accuracy increases as well in most cases. We demonstrate that the threshold can be learned reliably from a very small number of labeled examples, and supervised classifiers learned on such small datasets produce poorer bias and accuracy performances.