Nandana Nagabhushana and Natarajan S, PESIT, India

Data Mining is being increasingly used in the field of automation of decision making processes, which involve extraction and discovery of information hidden in large volumes of collected data. Nonetheless, there are negative perceptions like privacy invasion and potential discrimination which contribute as hindrances to the use of data mining methodologies in software systems employing automated decision making. Loan granting, Employment, Insurance Premium calculation, Admissions in Educational Institutions etc., can make use of data mining to effectively prevent human biases pertaining to certain attributes like gender, nationality, race etc. in critical decision making. The proposed methodology prevents discriminatory rules ensuing due to the presence of certain information regarding sensitive discriminatory attributes in the data itself. Two aspects of novelty in the proposal are, first, the rule mining technique based on ontologies and the second, concerning generalization and transformation of the mined rules that are quantized as discriminatory, into non-discriminatory ones

Ontology, Discrimination Prevention, Rule Protection, Rule Generalization, Postmining