Towards Building Child-Centered Machine Learning Pipelines - Use Cases from K-12 and Higher-Education

Aug 3, 2026

Upload by Logic

Human-centered guidance has emerged for building machine-learning (ML) pipelines, yet child-centered adaptation remains underexplored as children increasingly use ML-powered products. This paper articulates the key questions used to translate human-centered ML pipeline design into child-centered practice, and presents two end-to-end case studies for children’s products. It discusses the child-centered goals across pipeline stages, including problem identification, data collection, exploratory analysis, model development, ethics, and interface development, while emphasizing affordability and ethical considerations for K-12 and higher education contexts.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages5

Reading Time13 min

AccessFree to download

ML Pipeline Stages from a Child-Centered Perspective

Problem Identification

Data Collection

Exploratory Analysis

Model Development

Ethics Considerations

Interface Development

Case Study 1: Classroom Engagement Prediction

Problem Identification

arXiv :2304 .09532v1 [ cs .HC] 19 Apr 2023 Towards Building Child-Centered Machine Learning Pipelines: Use Cases from K-12 and Higher-Education ALPAY SABUNCUOĞLU, UNVEST R&D Center, Turkey CEYLAN BEŞEVLI, University College London (UCL), United Kingdom T. METIN SEZGIN, Koc University, Turkey Researchers and policy-makers have started creating frameworks and guidelines for building machine-learning* (ML) pipelines with a human-centered lens [1, 5]. Machine Learning pipelines stand for all the necessary steps to develop ML systems (e.g., developing a predictive keyboard) [7] . On the other hand, a child-centered focus in developing ML systems has been recently gaining interest as children are becoming users of these products [14] . These eﬀorts dominantly focus on children’s interaction with ML-based systems. However, from our experience, ML pipelines are yet to be adapted using a child-centered lens. In this paper, we list the questions we ask ourselves in adapting human-centered ML pipelines to child-centered ones. We also summarize two case studies of building end-to-end ML pipelines for children’s products. *Throughout the paper, we used the term ML rather than AI. Commercial examples use the term AI loosely to deﬁne all the systems that learn from data. However, the term AI covers a broad range of intelligence levels, from narrow intelligence to general intelligence, that tries to achieve systems that think and act humanly and rationally [8]. Current intelligent systems (including our work) are examples of narrow intelligence that use ML techniques to learn from data. Thus, we used the term ML throughout the paper. CCS Concepts: • Human-centered computing → Interactive systems and tools; Interface design prototyping; • Social and professional topics → K-12 education. Additional Key Words and Phrases: child-centered machine learning ACM Reference Format: Alpay Sabuncuoğlu, Ceylan Beşevli, and T. Metin Sezgin. 2023. Towards Building Child-Centered Machine Learning Pipelines: Use Cases from K-12 and Higher-Education. In CHI 2023 Workshop on Child-centred AI Design: Deﬁnition, Operation and Considerations, April 23, 2023, Hamburg, Germany. ACM, New York, NY, USA, 5 pages. 1 ML PIPELINES FROM A CHILD-CENTERED PERSPECTIVE The stages of a typical machine-learning pipeline can be listed in six main steps, as there’s no standardization yet [6] . Following these stages with a human-centered focus can help practitioners build natural, intuitive-to-use, highperformance intelligent systems that abide by ethical considerations. In our work, we both used pre-trained, ready-todeploy ML models and developed new models starting from scratch to increase the aﬀordability and potential outreach of our systems. Table 1 lists the steps and the corresponding questions we ask in the pipeline steps to make these projects more child-centered. Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for proﬁt or commercial advantage and that copies bear this notice and the full citation on the ﬁrst page. Copyrights for third-party components of this work must be honored. For all other uses, contact the owner/author(s) . © 2023 Copyright held by the owner/author(s) . Manuscript submitted to ACM CCAI 2023, April 23, 2023, Hamburg, Germany Alpay Sabuncuoğlu, Ceylan Beşevli, and T. Metin Sezgin | ML Pipeline Stages | Child-Centered Questions | | --- | --- | | Problem identiﬁcation | -Is the problem focusing on data collection, model development, or interaction?
-Are there any child-centered use cases or guidelines for this problem? | | Data Collection | -If there is an already existing dataset for the identiﬁed problem: How can we update it to make it more child-centered?
- If no dataset is available: What should be the steps for involving children in the data collection process?
-If it is a supervi

FAQs

What is the main goal of adapting human-centered ML pipelines to a child-centered lens?

To make the pipeline development process address questions specific to children as end users, covering the full workflow from data collection to interface design while considering ethical impacts on children.

Which stages does the paper use to structure child-centered ML pipeline thinking?

It organizes guidance into six steps: problem identification, data collection, exploratory analysis, model development, ethics considerations, and interface development.

What does Case Study 1 focus on and how is it implemented end-to-end?

It targets predicting classroom engagement during group activities by collecting data, building baseline architectures, and creating an interactive dashboard to present engagement levels in a student-centric way.

Aug 3, 2026

Upload by Logic