Article: Causal transformer and counterfactual reasoning: deconstruction analysis of the impact of teaching strategies on academic achievement Journal: International Journal of Computational Systems Engineering (IJCSYSE) 2025 Vol.9 No.17 pp.1 - 24 Abstract: Modern education increasingly relies on data-driven decision-making, requiring causal inference methods to assess teaching strategies beyond correlations. Challenges such as time-varying confounders, unobserved counterfactuals, temporal dependencies of interventions, and heterogeneous responses limit strategy design and evaluation. Existing methods also struggle with temporal dynamics and complex causal structures. To address these issues, causal temporal contextual reasoning (CTCR) was developed, incorporating a dynamic disentanglement mechanism for time-varying confounders, a two-way causal representation module, and a counterfactual generation algorithm constrained by temporal logic. Experiments on higher education (Dataset-H), K12, and MOOCs datasets show CTCR's effectiveness. On Dataset-H, it achieves MSE-T 5.53 × 10-2, PEHE 5.16 × 10-1, and CP@K 0.89, outperforming comparative models. Performance volatility across K12 and MOOCs is ≤ 29.3%, and CTCR remains robust under 30% data sparsity and 5 dB noise, demonstrating strong generalisation and reliability. Inderscience Publishers - linking academia, business and industry through research

Title: Causal transformer and counterfactual reasoning: deconstruction analysis of the impact of teaching strategies on academic achievement

Authors: Lipeng Zhang

Addresses: School of Marxism, Shijiazhuang College of Applied Technology, Shijiazhuang City, Hebei Province, 050800, China

Abstract: Modern education increasingly relies on data-driven decision-making, requiring causal inference methods to assess teaching strategies beyond correlations. Challenges such as time-varying confounders, unobserved counterfactuals, temporal dependencies of interventions, and heterogeneous responses limit strategy design and evaluation. Existing methods also struggle with temporal dynamics and complex causal structures. To address these issues, causal temporal contextual reasoning (CTCR) was developed, incorporating a dynamic disentanglement mechanism for time-varying confounders, a two-way causal representation module, and a counterfactual generation algorithm constrained by temporal logic. Experiments on higher education (Dataset-H), K12, and MOOCs datasets show CTCR's effectiveness. On Dataset-H, it achieves MSE-T 5.53 × 10^-2, PEHE 5.16 × 10^-1, and CP@K 0.89, outperforming comparative models. Performance volatility across K12 and MOOCs is ≤ 29.3%, and CTCR remains robust under 30% data sparsity and 5 dB noise, demonstrating strong generalisation and reliability.

Keywords: causal transformer; counterfactual reasoning; instructional strategy; CTCR; academic achievement; time-varying confounders.

DOI: 10.1504/IJCSYSE.2025.150995

International Journal of Computational Systems Engineering, 2025 Vol.9 No.17, pp.1 - 24

Received: 12 Aug 2025
Accepted: 23 Oct 2025
Published online: 07 Jan 2026 *

Title: Causal transformer and counterfactual reasoning: deconstruction analysis of the impact of teaching strategies on academic achievement

Keep up-to-date