Detection of outlying proportions

TL;DR

This paper presents a new efficient method for detecting outliers in proportions using contingency tables and specialized algorithms, with demonstrated effectiveness on synthetic and biological data.

Contribution

Introduces a novel outlier detection method for proportions based on contingency tables and tailored algorithms, with theoretical insights and practical demonstrations.

Findings

Effective outlier detection demonstrated on synthetic data

Good performance shown on biological experiment data

Algorithm efficiency improved by exploiting contingency table structure

Abstract

In this paper we introduce a new method for detecting outliers in a set of proportions. It is based on the construction of a suitable two-way contingency table and on the application of an algorithm for the detection of outlying cells in such table. We exploit the special structure of the relevant contingency table to increase the efficiency of the method. The main properties of our algorithm, together with a guide for the choice of the parameters, are investigated through simulations, and in simple cases some theoretical justifications are provided. Several examples on synthetic data and an example based on pseudo-real data from biological experiments demonstrate the good performances of our algorithm.