A unified approach to outlier identification for mixed-type continuous/ordinal data

arXiv:2606.26324 · stat.ME · Submitted 2026-06-24 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Statistics. The day in statistics: methods, theory and the inference under them.

Tom: I'm Tom.

Jane: And I'm Jane, here with Lu, Meng and our in-house large language model, Lalam.

Tom: We're looking at the paper: "A unified approach to outlier identification for mixed-type continuous/ordinal data".

Jane: Here's the gist of the paper. We present an outlier identification method for mixed type data sets comprising continuous and ordinal variables <ref:2606.26324#pg2>.

Tom: This approach is significant because it extends robust methods like the Minimum Covariance Determinant estimator to handle complex, real-world datasets that combine different data types, thereby improving the ability to detect anomalies in such mixed contexts <ref:2606.26324#pg2>.

Jane: Problem Context Real world data often contain outliers <ref:2606.26324#pg2>.

Tom: Outliers can severely undermine statistical analyses, and their detection may also be of interest in its own right <ref:2606.26324#pg2>.

Paper summary: Tom: So to wrap up, the paper "A unified approach to outlier identification for mixed-type continuous/ordinal data" presents an extension of the Minimum Covariance Determinant estimator designed specifically for datasets that mix continuous and ordinal variables.

Jane: The core thesis is using a multivariate Gaussian distribution as a reference for non-outliers while assuming a latent Gaussian structure under the hood for those ordinal variables, which lets them define outliers consistently across both types of data.

Lu: They achieve this by extending MCD using notions like polychoric and polyserial correlations to robustly estimate the parameters of that reference distribution, even though they don't see the full Gaussian information underlying the ordinal variables.

Meng: The paper shows simulations on synthetic data with various contamination levels, confirming high detection accuracy with a low false positive rate, which suggests this method is empirically sound for finding anomalies in this mixed context.

Lalam: The application on the London Airbnb data set provided specific insights into how these outliers are distributed geographically and linked them to economic factors like advertised prices in central locations.

Tom: Ultimately, the implication is that we now have a more robust tool available for identifying unusual patterns in complex, real-world datasets that contain both continuous measurements and ordered categories.

Jane: It’s about providing a single mathematical approach instead of needing separate tools for handling continuous data versus ordinal data when they live together in the same analysis.

Conclusion: Tom: So we're wrapping up this look at this paper by checking out how they unified outlier detection for continuous and ordinal data sets using the Minimum Covariance Determinant estimator, or MCD.

Jane: It sounds like they took a problem where you have both numbers and categories—like a price in dollars mixed with an ordered rating—and created one single robust method to find the weird stuff across both types at the same time.

Tom: Exactly. The authors are trying to solve this mess where we usually have separate tools for continuous data and separate tools for categorical data, but they're proposing one framework based on a shared Gaussian reference distribution.

Lu: What’s interesting is how they handle that ordinal part by assuming it comes from some underlying latent Gaussian structure, which lets them use the same core outlier definition principle across everything.

Meng: From an engineering standpoint, the challenge must have been estimating those correlation matrices robustly enough to define the scale and location parameters for both types simultaneously without getting swamped by noise.

Lalam: I'm seeing that their simulation results show they can find true outliers with a low false positive rate, which is what we need when we're actually trying to spot genuine anomalies in production data.

Jane: So, it’s not just a theoretical tweak; it’s about making anomaly detection practical for the messy, mixed real-world data that most of us deal with every day.

Tom: Right. It moves the goal from building two separate systems to building one system that works on mixed data and actually delivers reliable results when you deploy it.

Jane: And thinking about what this means for future work, we need to see how they handle things like very high dimensions or datasets where the underlying structure isn't even close to a simple Gaussian assumption.

Efthymios Costa, Christian Hennig

Imperial College London · University of Bologna

stat.ME

Submitted: 2026-06-24

Updated: 2026-11-03

Comments: 30 pages

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 77/100

The gist: We present an outlier identification method for mixed type data sets comprising continuous and ordinal variables <ref:2606.26324#pg2>.

Key concepts

Mixed Type Data Sets
These are datasets containing different kinds of variables, specifically continuous numbers and ordinal categories. The paper addresses the challenge that existing outlier detection methods often only work well for one type of data, making it hard to find anomalies when both types coexist.
Minimum Covariance Determinant (MCD) Estimator
MCD is a robust statistical technique used to find the 'center' or 'normal' distribution of a dataset while ignoring extreme outliers. The paper adapts this robust principle to handle the complexity introduced by combining continuous and ordinal variables.
Latent Gaussian Assumption
This concept models how ordinal variables might have been generated. It assumes that even though we only see ordered categories, they originated from an underlying, unobserved continuous Gaussian distribution. This allows the method to treat them similarly to continuous data for outlier detection purposes.

Terminology

Summary

We present an outlier identification method for mixed type data sets comprising continuous and ordinal variables <ref:2606. p. 0

Where the paper says it

hennig@unibo. it, ORCID 0000-0003-1550-5637 June 30, 2026 Abstract We present an outlier identification method for mixed type data sets comprising continuous and ordinal variables.

26324#pg2>. This approach is significant because it extends robust methods like the Minimum Covariance Determinant estimator to handle complex, real-world datasets that combine different data types, thereby improving the ability to detect anomalies in such mixed contexts <ref:2606. p. 1
Where the paper says it

Techniques that can handle such data and are robust to outliers are scarce. In this paper, we extend the Minimum Covariance Determinant estimator (MCD; Rousseeuw (1984)) to the mixed continuous-ordinal setting, and we use it to identify outliers.

26324#pg2>.

Problem Context

Real world data often contain outliers <ref:2606.26324#pg2>. Outliers can severely undermine statistical analyses, and their detection may also be of interest in its own right <ref:2606. p. 1

Where the paper says it

1 Introduction Real world data often contain outliers. Outliers can severely undermine statistical analyses, and their detection may also be of interest in its own right.

26324#pg2>. While robust methods exist for purely continuous data, many applications involve variables of mixed types, combining continuous and categorical features <ref:2606. p. 1
Where the paper says it

Outliers can severely undermine statistical analyses, and their detection may also be of interest in its own right. While robust methods exist for purely continuous data, many applications involve variables of mixed types, combining continuous and categorical features.

26324#pg2>. Techniques that can handle such data and are robust to outliers are scarce <ref:2606. p. 1
Where the paper says it

While robust methods exist for purely continuous data, many applications involve variables of mixed types, combining continuous and categorical features. Techniques that can handle such data and are robust to outliers are scarce.

26324#pg2>. Ordinal variables as considered here have a finite number of possible outcome categories, which are ordered <ref:2606. p. 1
Where the paper says it

Ordinal variables as considered here have a finite number of possible outcome categories. These categories are ordered.

26324#pg2>. The ordinality of such variables is defined by the absence of a meaningful numerical distance between any two categories <ref:2606. p. 1
Where the paper says it

These categories are ordered. Ordinality of such variables is defined by the absence of a meaningful numerical distance between any two categories (Stevens 1946).

26324#pg2>.

Proposed Methodology

The proposed algorithm is based on using a multivariate Gaussian distribution as a reference distribution for non-outliers, with a latent Gaussian assumed for ordinal variables <ref:2606.26324#pg2>. This method extends the Minimum Covariance Determinant estimator (MCD) to the mixed continuous/ordinal setting <ref:2606. p. 1

Where the paper says it

Techniques that can handle such data and are robust to outliers are scarce. In this paper, we extend the Minimum Covariance Determinant estimator (MCD; Rousseeuw (1984)) to the mixed continuous-ordinal setting, and we use it to identify outliers.

26324#pg2>. The core idea is to model ordinal variables as having been generated from an underlying unobserved latent Gaussian (“ordinal-LG” in the following) <ref:2606. p. 1
Where the paper says it

To our knowledge there is currently no version of the MCD that extends to mixed type data. Here we generalize MCD-based outlier identification to mixed type continuous/ordinal data, assuming a multivariate Gaussian distribution for a majority of non-outlying data where ordinal variables are modeled as having been generated from an underlying unobserved latent Gaussian (“ordinal-LG” in the following).

26324#pg2>. This unification of outlier definition based on a multivariate Gaussian reference distribution for continuous and ordinal variables is the central proposal <ref:2606. p. 3
Where the paper says it

3 Outliers based on the latent Gaussian approach for mixed type data 3. 1 Outlier region for a multivariate Gaussian reference distribution After reviewing the definition of outliers given by Becker & Gather (1999) for a multivariate Gaussian reference distribution for continuous data, we extend this definition to mixed type continuous/ordinal data.

26324#pg2>.

Estimation and Modeling

To estimate the outlier regions, it is necessary to estimate the parameters µ and Σ of X˜ in a robust way <ref:2606.26324#pg2>. This requires estimating entries of Σ corresponding to pairs of continuous variables, pairs of ordinal variables, and pairs with one continuous and one ordinal variable <ref:2606. p. 7

Where the paper says it

This is done by an adaptation of the MCD principle. For mixed type data, this requires estimating entries of Σ that correspond to pairs of continuous variables, pairs of ordinal variables, and pairs with one continuous and one ordinal variable.

26324#pg2>. This is achieved using notions of polychoric and polyserial correlations <ref:2606. p. 7
Where the paper says it

For mixed type data, this requires estimating entries of Σ that correspond to pairs of continuous variables, pairs of ordinal variables, and pairs with one continuous and one ordinal variable. In order to do this, we use the notions of polychoric and polyserial correlations (Pearson 1900, 1909).

26324#pg2>. The MCD principle can be applied to estimating correlations rather than covariances by computing correlations on the h-subset <ref:2606. p. 8
Where the paper says it

The MCD principle can be applied to estimating correlations rather than covariances by computing correlations on the h-subset. These can be computed as plain Pearson correlations between two continuous variables, polychoric correlation between two ordinal variables, and polyserial correlation between a continuous and an ordinal variable.

26324#pg2>. The continuous/ordinal mixed type MCD estimator of location and scale is defined using estimated thresholds and robust correlation matrices <ref:2606. p. 9
Where the paper says it

3 Definition of the estimator Given a data set Xn, using the notation for data sets from Section 3. 2, the continuous/ordinal mixed type MCD estimator of location and scale is defined as follows (we will later use the term MCD for this, and notation connected to the MCD such as HMCD will be used referring to the mixed type variables case from now on).

26324#pg2>.

Robustness and Validation

The method's robustness is demonstrated via simulations on synthetic data sets with various types of contamination, achieving high detection and low false positive rates <ref:2606.26324#pg2>. The breakdown theorem shows that replacing observations will not stop extreme enough outliers from being identified <ref:2606. p. 0

Where the paper says it

This is extended to account for the fact that the full Gaussian information underlying the ordinal variables is not observed. A breakdown theorem shows that replacing observations will noty stop extreme enough outliers from being identified.

26324#pg2>. In simulations, the MCD-omix achieves high outlier detection accuracy, effectively identifying true outliers with a low false-positive rate <ref:2606. p. 0
Where the paper says it

A breakdown theorem shows that replacing observations will noty stop extreme enough outliers from being identified. The effectiveness of our approach is demonstrated via simulations on synthetic data with various types of contamination, achieving high detection and low false positive rates.

26324#pg2>. Practical relevance is illustrated through an application to Airbnb listing data containing both continuous and ordinal attributes <ref:2606. p. 0
Where the paper says it

The effectiveness of our approach is demonstrated via simulations on synthetic data with various types of contamination, achieving high detection and low false positive rates. Practical relevance is illustrated through an application to Airbnb listing data containing both continuous and ordinal attributes.

26324#pg2>.

Application Insights

The application on the London Airbnb data set revealed associations not identified by spatial models <ref:2606.26324#pg2>. The method detected 33 observations, representing just 0. p. 2

Where the paper says it

In Section 5 we present a theorem that makes sure that, given an original dataset and using our method, extreme enough outliers w. this dataset are still identified if a certain maximum number of observations are arbitrarily replaced.

7% of the listings in the data set <ref:2606. p. 2
Where the paper says it

We present some simulations on synthetic data sets with no outliers and with four different types of contamination in Section 6. The proposed approach is applied on a data set of Airbnb listings from London during the weekdays of 2021 in Section 7.

26324#pg2>. Outlying listings were evenly spread north and south of the Thames river, with Greenwich being the borough including the most outliers <ref:2606.26324#pg2>. The analysis provided insights regarding advertised listings, identifying instances that can be considered “tourist traps” as they rely on first time travellers’ ignorance to set unreasonably high prices for centrally located listings <ref:2606.26324#pg2>.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Outlier Detection in Mixed Data Systems: The MCD-omix method can be integrated into anomaly detection pipelines for datasets containing both continuous and ordinal features, providing a unified framework based on robust Mahalanobis distances that account for latent Gaussian assumptions in ordinal variables.

  2. Robust Parameter Estimation: The system can utilize the MCD estimator to robustly estimate the location and scatter of continuous variables while employing polychoric and polyserial correlations (estimated using maximum likelihood conditional on estimated thresholds) to model the dependence structure involving ordinal variables.

  3. Adaptive Outlier Thresholding: By employing a breakdown theorem, the system can be designed such that extreme enough outliers w.r.t this dataset are still identified if a certain maximum number of observations are arbitrarily replaced, ensuring high detection accuracy even under substantial contamination.

  4. Feature Space Mapping: The output Mahalanobis distances provide a measure of how far an observation lies from the estimated non-outlier distribution, allowing the AI system to map and visualize atypical feature combinations, such as high cleanliness ratings but low overall guest satisfaction scores.

Abstract

We present an outlier identification method for mixed type data sets comprising continuous and ordinal variables. We define outliers based on using a multivariate Gaussian distribution as reference distribution for non-outliers, with a latent Gaussian assumed for ordinal variables. The proposed algorithm is based on the robust Minimum Covariance Determinant estimator for estimating the parameters of the multivariate Gaussian for the non-outliers. This is extended to account for the fact that the full Gaussian information underlying the ordinal variables is not observed. A breakdown theorem shows that replacing observations will noty stop extreme enough outliers from being identified. The effectiveness of our approach is demonstrated via simulations on synthetic data with various types of contamination, achieving high detection and low false positive rates. Practical relevance is illustrated through an application to Airbnb listing data containing both continuous and ordinal attributes.

Related papers