Data preprocessing approach for removal of direct and indirect discrimination

Abstract: Data mining is an important technology for extract the useful knowledge hidden in large collection of data. Automated data collection and data mining techniques such as classification rule mining have proved the way to making automated decisions, like loan granting, insurance premium computation, etc. If the training data set is biased in what regards sensitive attributes like gender, race, religion, colour, etc., . For this reason, anti-discrimination techniques including discrimination discovery and prevention have been introduced in data mining. Discrimination can be direct or indirect. Direct discrimination occurs only when decisions are made based on their sensitive attributes where indirect discrimination occurs when decisions are made based on their no sensitive attributes which are strongly related with biased sensitive ones. In this paper, we focus discrimination prevention in data mining and propose a new techniques applicable for direct or indirect discrimination prevention individually or both at the same time. Also we discuss how to clean training data set and outsourced data sets in such a way that direct and indirect discriminatory decision rules are converted to legitimate (non-discriminatory) classification rules. We also propose new metrics to evaluated the utility of the proposed approach and we compare these approach. The experimental evaluations implemented that the proposed techniques are effective removed by direct and/or indirect discrimination biases in the original data set while preserving data quality.

Keywords: Data mining, that direct and indirect discriminatory decision rules