Threshold-like Structure and Barrier–Benefit Asymmetry in Students’ Behavioural Intention Toward E-Learning: An Explainable Machine Learning Analysis in Post-Pandemic Higher Education

This study examines students’ behavioural intention (BI) to use e-learning in post-pandemic higher education, addressing a critical research gap: the tendency of standard linear models to obscure the complex, hidden structural mechanisms of technology acceptance. Building on the Technology Acceptance Model (TAM), the article re-analyses post-COVID-19 survey data from a private university in Warsaw, Poland, to test whether behavioural intention is shaped only by linear effects or also by hidden nonlinear mechanisms. Complementary explainable machine learning models (XAIs) are used to identify the model’s stable linear core, threshold-like structure, and nonlinear predictor effects. The findings confirm the central role of perceived usefulness (PU) and perceived ease of use (PEOU) within the TAM framework. Crucially, the results show that, once the broader attitudinal domain toward remote learning is decomposed into negative (NEG) and positive (POS) evaluative components, negative evaluations exert a substantially stronger suppressive effect on behavioural intention than positive evaluations exert as facilitators. In the standardised linear benchmark, the absolute effect of NEG is nearly twice that of POS, with SHapley Additive exPlanations (SHAP) analysis confirming this asymmetry at the nonlinear level. Furthermore, nonlinear models (partial dependence plots and decision trees) identify a threshold-like transition region in the PU-BI relationship, where acceptance increases rapidly only after exceeding a specific utility level. For educational policy, these results imply a hierarchical intervention strategy: institutional efforts should prioritise crossing the critical utility threshold and focus on systematically reducing specific barriers (NEG) rather than relying solely on promoting the general benefits of digital formats. Overall, the study confirms the TAM as a robust baseline framework while showing that post-pandemic e-learning acceptance is additionally shaped by asymmetric and threshold-like mechanisms that remain partly hidden in standard linear models.