Evidential likelihood flatness as a way to measure data quality: the multinomial case

Abstract : Likelihood functions, as well as the more recent concept of evidential likelihood, are essential statistical tools to perform estimation. Beyond the maximal likelihood value, the shape of the likelihood can also give interesting information about the data used to get the estimation. Indeed, it is generally acknowledged that more uncertain and scarce data will lead to flatter likelihoods. However, different aspects can affect this shape, and it may be worthwhile to separate various influences. In this paper, we discuss these influences and propose some practical ways to separate them into different measures. We demonstrate our approach on the particular case of the multinomial likelihood, which plays an important role in applications such as classification.
Document type :
Conference papers
Complete list of metadatas

Cited literature [1 references]  Display  Hide  Download

Contributor : Sébastien Destercke <>
Submitted on : Tuesday, January 12, 2016 - 9:33:38 AM
Last modification on : Tuesday, July 24, 2018 - 4:40:02 PM
Long-term archiving on : Friday, April 15, 2016 - 9:40:35 PM


Files produced by the author(s)


  • HAL Id : hal-01254287, version 1



Liyao Ma, Sébastien Destercke, Yong Wang. Evidential likelihood flatness as a way to measure data quality: the multinomial case. 16th World Congress of the International Fuzzy Systems Association and the 9th Conference of the European Society for Fuzzy Logic and Technology (IFSA-EUSFLAT 2015), Jun 2015, Gijon, Spain. pp.313-319. ⟨hal-01254287⟩



Record views


Files downloads