Explainable machine learning for predicting self-reported digital competence levels among university faculty

Introduction Digital competence has become an important component of university faculty practice across teaching, research, outreach, and academic management. Explainable machine learning provides an analytical approach for classifying self-reported digital competence levels and examining the relative contribution of sociodemographic, academic, and institutional predictors. The objective of this study was to predict self-reported digital competence levels among university faculty at the Technical University of Manabí and the State University of Milagro using explainable machine learning. Methods A quantitative, non-experimental, cross-sectional design with a predictive classification approach was used. Data were collected from 315 faculty members using a 300-item Likert-type self-report questionnaire covering digital competence in teaching, research, outreach, and academic management. The outcome was operationalized as self-reported digital competence and categorized into low, medium, and high levels using percentile thresholds estimated from the questionnaire scores within the analytical sample. These categories therefore represent relative positions within the observed self-report distribution rather than externally validated or performance-based competence standards. Predictive performance was evaluated using stratified 10-fold cross-validation, with each observation predicted only when it was held out from the fold used to fit that model. Results Pooled out-of-fold predictions were used for the confusion matrices and do not constitute an independent external test set. Several supervised machine learning algorithms demonstrated high internal discriminative performance in classifying low, medium, and high levels of self-reported digital competence. Model performance varied across algorithms, while explainability analyses identified the relative contribution of sociodemographic, academic, professional, and institutional predictors to the classification outcome. Conclusions Explainable machine learning showed substantial within-sample capacity to classify questionnaire-derived levels of self-reported digital competence. Because the low, medium, and high categories were defined using percentile thresholds derived from the same self-report instrument and analytical sample, they should be interpreted as relative classifications within the observed distribution rather than as externally validated standards of digital competence.