Genetic learning of fuzzy integrals accumulating human-reported environmental stress

In this paper, we develop models based on fuzzy integrals (both of the Choquet and Sugeno type) for accumulating annoyance by noise, odor or light caused by particular sources or activities. As underlying fuzzy measures, we have opted for k-maxitive measures (in particular 1-maxitive or 2-maxitive) as the best known crisp model points in this direction. The fuzzy measures are learnt from survey data and optimized using genetic algorithms. Attention is paid to several types of inconsistencies that typically arise in data sets collected through social surveys. Also, special care is taken to make sure that the Sugeno integral and the genetic algorithm that optimizes the associated fuzzy measure operates solely on the ordinal scale of linguistic labels.