Uniting Mcdonald’s Beta and Liouville Distributions to Empower Anomaly Detection

Oussama Sghaier, Manar Amayri, Nizar Bouguila

2025

Abstract

In this paper, we examine the McDonald’s Beta-Liouville distribution, a new distribution that combines the key features of the Liouville and McDonald’s Beta distributions, in order to address the issue of anomaly identification in proportional data. Its primary advantages over the standard distributions for proportional data, including the Dirichlet and Beta-Liouville, are its flexibility and capacity for explanation when working with this type of data, thanks to its variety of presented parameters. We provide two discriminative methods: a feature mapping approach to improve Support Vector Machine (SVM) and normality scores based on choosing a specific distribution to approximate the softmax output vector of a deep classifier. We illustrate the advantages of the proposed methods with several tests on image and non-image data sets. The findings show that the suggested anomaly detectors, which are based on the McDonald’s Beta-Liouville distribution, perform better than baseline methods and classical distributions.

Paper Citation

in Harvard Style

Sghaier O., Amayri M. and Bouguila N. (2025). Uniting Mcdonald’s Beta and Liouville Distributions to Empower Anomaly Detection. In Proceedings of the 27th International Conference on Enterprise Information Systems - Volume 1: ICEIS; ISBN 978-989-758-749-8, SciTePress, pages 181-190. DOI: 10.5220/0013164100003929

in Bibtex Style

@conference{iceis25,
author={Oussama Sghaier and Manar Amayri and Nizar Bouguila},
title={Uniting Mcdonald’s Beta and Liouville Distributions to Empower Anomaly Detection},
booktitle={Proceedings of the 27th International Conference on Enterprise Information Systems - Volume 1: ICEIS},
year={2025},
pages={181-190},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013164100003929},
isbn={978-989-758-749-8},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 27th International Conference on Enterprise Information Systems - Volume 1: ICEIS
TI - Uniting Mcdonald’s Beta and Liouville Distributions to Empower Anomaly Detection
SN - 978-989-758-749-8
AU - Sghaier O.
AU - Amayri M.
AU - Bouguila N.
PY - 2025
SP - 181
EP - 190
DO - 10.5220/0013164100003929
PB - SciTePress