— Digital technologies are acknowledged as game-changers for higher education, making information-system acceptance essential to the successful implementation of digital technologies. Nevertheless, present-day acceptance research mostly employs traditional statistical ones or single machine-learning models that may not efficiently model the intricate nonlinear differentiation amongst acceptance factors. In this study, we tackle this issue with a method including Deep Ensemble consists of MLP-Multiple Layer Perceptron, DNN-Deep Neural Network, and Wide & Deep architectures in addition to TabNetfor information-system acceptance prediction. Our approach is novel in marrying combinations of complementary deep-learning configurations with feature selection and the use of a meta-learner to improve predictive power and generalisability. The proposed framework was evaluated using survey-based higher-education data through classification, continuous prediction, and 5-fold cross-validation. Results demonstrate strong predictive performance, achieving 97.42% accuracy, 99.50% ROC-AUC, 99.10% PR-AUC, and 97.40% F1-score, while the model achieved 96.87% ± 0.72% cross-validation accuracy. The findings demonstrate that the proposed ensemble provides a reliable approach for predicting information-system acceptance in higher education. Digital information systems are reshaping higher education, yet their benefits materialize only when students, teachers and administrators accept them. Acceptance research still relies mainly on correlational statistics or single machine-learning models, which cannot capture the nonlinear, interacting and heterogeneous structure of technology-acceptance factors. This paper proposes an intelligent computing framework, a heterogeneous Deep Ensemble that fuses four deep architectures with deliberately different inductive biases: a multilayer perceptron (MLP) for smooth nonlinear mappings, a deep neural network (DNN) for hierarchical representations, a Wide & Deep network for joint memorization and generalization of feature interactions, and TabNet for attention-driven sparse feature selection on tabular data. Their class-probability outputs are fused by a trainable neural meta-learner, which turns architectural diversity into a single calibrated acceptance predictor. The framework integrates preprocessing, correlation and multicollinearity analysis and feature selection, and it is validated through classification, discrimination, probabilistic and continuous-prediction metrics together with 5-fold cross-validation. On survey data from higher education, it achieved 97.42% accuracy, 97.40% F1-score, 99.50% ROC-AUC, 99.10% PR-AUC, 0.949 MCC and a Brier score of 0.018, with 96.87% ± 0.72% cross-validation accuracy. These results show that diversity-aware deep ensemble intelligence is an accurate, well-calibrated and stable approach to AI-driven decision support for digital transformation in education.

Information system acceptance; Deep ensemble learning; Higher education; Technology adoption Artificial intelligence; deep ensemble learning; meta-learning; TabNet; Wide & Deep; intelligent computing; information system acceptance; higher education