Résumé
Histograms are very useful for summarizing statistical information associated with a set of observed data. They are one of the most frequently used density estimators due to their ease of implementation and interpretation. However, histograms suffer from a high sensitivity to the choice of both reference interval and bin width. This paper addresses this difficulty by means of a fuzzy partition. We propose a new density estimator based on transferring the counts associated with each cell of the fuzzy partition to any subset of the reference interval. We introduce three different methods of achieving this transfer. The properties of each method are illustrated with a classic real observation set. The density estimator obtained relates to the Parzen–Rosenblatt kernel density estimation technique. In this paper, we only consider the monovariate case with precise and imprecise observations.