A METHODOLOGY FOR DIRECT AND INDIRECT DISCRIMINATION PREVENTION IN DATA MINING

Abstract

Along with privacy, discrimination is a very important issue when considering the legal and ethical aspects of data mining. It is more than observable that the majority people do not want to be discriminated because of their gender, nationality, religion, age and so on, particularly when those aspects are used for making decisions about them like giving them a occupation, loan, insurance, etc. determining such possible biases and eliminating them from the training data without harming their decision-making utility is therefore extremely popular. For this reason, antidiscrimination methods containing discrimination detection and prevention have been introduced in data mining. Discrimination prevention consists of suggest models that do not lead to discriminatory decisions even if the original training datasets are essentially biased. In this section, by focusing on the discrimination prevention, we present taxonomy for classifying and examining discrimination prevention schemes. Then, we begin a group of pre-processing discrimination prevention schemes and indicate the special features of each approach and how these approaches deal with direct or indirect discrimination. A production of metrics used to estimate the performance of those approaches is also specified. In conclusion, we finish our learn by specifying interesting future directions in this research body.