A METHOD OF ADJUSTING WEIGHTS IN MULTICRITERIA DECISION ANALYSIS BASED ON FAIRNESS MEASURES

In data analysis, the presence of bias in datasets have an important space in scientific discussions nowadays. In the ranking context, using Multi-Criteria Decision Analysis, the search of fair rankings, capable of mitigating biases effect involving specific groups of alternatives, can be done by adjusting the criteria weights considered in a certain problem, since the weights influence the order of the alternatives which influences the resulting ranking fairness mesure. In this study, we present a new weight adjustment method based on the optimization of the fairness measure known as NDKL (Normalized Discounted Cumulative KL-divergence). Our method, tested on synthetic and real data, consists of defining weights w that enable fair rankings based on the NDKL optimization in funtion of these weights.