The mode of interval-valued data

© 2019 IEEE. When applied to interval-valued data, averaging aggregation functions do not, in general, produce a good representative value of an interval data set. For real-valued scalar and vector data the mode has been shown to be both a robust averaging function, insensitive to noise and outliers, and to produce a good representative value of a larger set. This article introduces a method for computing the mode of sets of interval-valued data based on density estimation using the 2-additive Choquet integral over nearest neighbours on the unit simplex. This method is applied to compute the mode of daily temperature range data sets for Australian state capital cities. It is found that dissimilarity between the mode and the interval formed from averaging range end points independently increases significantly in months having highly skewed maximum or minimum temperature distributions.