One-class vs binary machine learning classification of ceramic samples described by chemical element concentrations

Aug 5, 2026

Upload by Quinn

Pottery provenance classification using chemical composition is widely supported by non-invasive analytical techniques, and machine learning increasingly enables data-driven archaeological classification. When provenance is treated as local and non-local evidence, one-class models can be more appropriate because they are trained only on positive examples and evaluated on both positive and negative data to assess generalization. The study tests a one-class SVM on 112 ceramic-fragment samples described by nine elements and compares it with binary learning models. Nested cross-validation and multiple metrics are used to quantify performance under assumptions about measurement and fragment-level dependence.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

AccessFree to download

Introduction

Pottery provenance and chemical characterization

Role of non-invasive analytical techniques (ED-XRF)

Problem setting: local vs non-local samples

Motivation for one-class models

Methodology

Dataset and feature representation (nine elements)

Data stratification to handle fragment/measure dependence

One-class SVM and binary classifiers

Nested cross validation and performance metrics

Results and discussion

Sensitivity and specificity comparison

Behavior on positive and negative examples

Journal of Cultural Heritage 71 (2025) 234–241 Contents lists available at ScienceDirect Journal of Cultural Heritage journal [homepage:](homepage: www.elsevier.com/locate/culher)[ www.elsevier.com/locate/culher](homepage: www.elsevier.com/locate/culher) | VSI:AI (methods) for cultural heritage
One-class vs binary machine learning classiﬁcation of ceramic samples described by chemical element concentrations
Dario Malchiodia,c, Anna Maria Zanabonia,c,∗, Alessandro Di Gioacchino a, Letizia Bonizzonib | | | | | --- | --- | --- | --- | | aDipartimento di Informatica, Università degli Studi di Milano, Via Celoria 18, Milano, 20133, Italy b Dipartimento di Fisica“Aldo Pontremoli”, Università degli Studi di Milano, Via Celoria 16, Milano, 20133, Italy c Data Science Research Center, Università degli Studi di Milano, Via Celoria 18, Milano, 20133, Italy | | | | | a r t i c l e i n f o | | a b s t r a c t | | | Article history:
Received 18 June 2024
Revised 24 October 2024
Accepted 19 November 2024 | | Pottery classiﬁcation based on chemical composition characterization through non-invasive analytical techniques is a well-known method typically adopted to solve the problem of pottery provenance attribution. Machine learning approaches have been recently introduced as a tool to develop models for archaeological classiﬁcation directly inferred from data. This classiﬁcation is quite often given in terms of local or non-local samples: in these cases, if the hypothesized provenance is to be validated, one-class models could be more adequate than binary or multi-class predictors. Indeed, one-class classiﬁers are trained only using positive examples of the class to be learned, and they can be subsequently tested on positive and negative examples in order to evaluate their generalization capability. They are thus in principle more apt to eﬃciently classify local samples, as the non-local ones do not naturally gather ina well-deﬁned class. In this paper, we tested a one-class classiﬁer on a dataset of 112 examples representing pottery fragments described in terms of nine chemical elements. Different examples of the dataset can correspond to measurements done on a same physical fragment, thus the hypothesis of independence among observations might be violated. For this reason, we investigated the use of three data stratiﬁcation techniques, based on physical fragments and on measures. We employed the support vector one-class classiﬁcation algorithm, ﬁnding the smallest sphere in a feature space that contains most of the training points. The obtained classiﬁcation performances were compared with those of several machine learning algorithms for binary classiﬁcation. All the models were trained by a nested cross validation technique, separately taking into account the ﬁne-tuning of hyperparameters and the robust estimation of generalization performance. Comparisons were done on the same dataset according to different performance metrics. The obtained results show that one-class classiﬁcation attains a similar sensitivity of binary classiﬁcation approaches, meanwhile improving the performance in terms of speciﬁcity, therefore showing a good behavior both on positive and negative examples.
© 2024 Published by Elsevier Masson SAS. | | | Keywords:
Machine learning
One-class classiﬁcation Ancient pottery classiﬁcation | | | | 1. Introduction Provenance classiﬁcation of the raw materials used for ancient ceramic production represents one of the most popular subjects in the ﬁeld of science-based archaeology, and it necessarily relies on interdisciplinary studies involving both the scientiﬁc and humanistic ﬁelds [1]. The determination of the chemical composition of ceramics is only one aspect, but it involves both the identiﬁcation of raw materials and the determination of ∗ Corresponding author. E-mail addresses: [dario.malchiodi@unimi.it](dario.malchiodi@unimi.it) (D. Malchiodi), [annamaria.zanaboni@unim

FAQs

Why are one-class classifiers considered for pottery provenance validation?

They are trained only on positive examples of the target class and can be tested against both positive and negative examples, making them suitable when the presumed provenance does not form a well-defined class for non-local samples.

What dataset and input description were used in the study?

The study uses 112 pottery-fragment examples, where each example is described by concentrations of nine chemical elements.

How did the authors address potential dependence between observations?

They examined three data stratification techniques based on physical fragments and on measures, since multiple measurements may correspond to the same physical fragment and violate independence assumptions.

Aug 5, 2026

Upload by Quinn