Predicting Faculty Engagement in Communities of Practice: A Neural Network Approach with Synthetic Data Augmentation

Communities of Practice (CoPs) offer valuable faculty development platforms in higher education, yet their evaluation is hindered by small sample sizes and predominantly descriptive analytics.This study addresses this methodological gap by developing a deep learning framework enhanced by data augmentation techniques to predict faculty engagement in CoPs.We collected survey data measuring six CoP dimensions from 29 academic staff at Arab Open University-Oman and expanded this limited dataset using Synthetic Minority Oversampling Technique (SMOTE), categorical permutation, and Gaussian noise injection.The resulting neural network model achieved 89% accuracy and an F1-score of 0.89, significantly outperforming both baseline logistic regression (69%) and non-augmented deep learning (82%) approaches.Feature importance analysis identified shared leadership and peer collaboration as the strongest predictors of engagement, while t-SNE visualizations revealed distinct behavioral clusters among faculty.This study contributes to educational research by demonstrating that (1) deep learning with appropriate data augmentation can effectively model faculty engagement patterns even with limited initial data, and (2) CoP theoretical constructs can be computationally validated and operationalized, enabling more targeted faculty development interventions.The framework provides institutions with a scalable method for evidence-based decision-making in professional development planning, helping identify disengaged faculty and prioritize high-impact support strategies.