Learning Analytics via Causal AI: Counterfactual Simulation of Academic Performance Using Socioeconomic Factors
IEEE Access · 2026 · 14 · 163734-163750
DOI: 10.1109/access.2026.3745305 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Paulo Quinsacara Jofré und 6 weiteren Autor:innen. Die Arbeit nennt 47 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Chilean National Agency for Research and Development (ANID), Fondecyt.
Abstract (Originalsprache)
Learning technologies increasingly collect large-scale student data, yet most educational analytics remain predictive and offer limited support for reasoning about possible interventions. This work proposes a counterfactual modeling framework that can serve as a decision-support layer for educational analytics. Using 270,411 student records from the 2021 Chilean Higher Education Access Test (PAES), we model the standardized high-school achievement score as the main academic outcome. The framework combines causal discovery, graph falsification, theory-guided validation, and counterfactual simulation. We apply PC and LiNGAM algorithms to infer directed acyclic graphs from observational educational data, assess their plausibility, and contrast the discovered relations with a theory-informed structural model. We then analyze intervention-oriented scenarios involving student labor participation and rural school location. Results show that student work is consistently associated with lower academic performance and, under the assumed causal model, a counterfactual intervention setting nonworking students to working status yields an estimated average reduction between 24.87 and 6.57 points in their scores. Rural school location shows a statistically significant but substantially smaller effect. These findings suggest that causal and counterfactual modeling can support interpretable what-if reasoning for educational decision-making. The proposed framework can operate as a transparent decision-support layer within learning analytics systems, enabling stakeholders to explore plausible intervention scenarios under explicit causal assumptions.
Förderung
- Chilean National Agency for Research and Development (ANID), Fondecyt
Zitieren
APA
Jofré, P. Q., Peralta, B., Schwarzenberg, P., Astudillo, H., Carvajal, A. P., Lévano, M., & Hidalgo, M. (2026). Learning Analytics via Causal AI: Counterfactual Simulation of Academic Performance Using Socioeconomic Factors. IEEE Access, 14, 163734-163750. https://doi.org/10.1109/access.2026.3745305
Vancouver
Jofré PQ, Peralta B, Schwarzenberg P, Astudillo H, Carvajal AP, Lévano M, et al. Learning Analytics via Causal AI: Counterfactual Simulation of Academic Performance Using Socioeconomic Factors. IEEE Access. 2026;14:163734-163750. doi:10.1109/access.2026.3745305
BibTeX
@article{Jofr2026,
author = {Jofré, Paulo Quinsacara and Peralta, Billy and Schwarzenberg, Pablo and Astudillo, Hernán and Carvajal, Alejandro Pérez and Lévano, Marcos and Hidalgo, Mauricio},
title = {Learning Analytics via Causal AI: Counterfactual Simulation of Academic Performance Using Socioeconomic Factors},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {163734--163750},
doi = {10.1109/access.2026.3745305}
}
EUVIMED