Cloud-Based Educational Data Mining and Adoption Prediction Using Ensemble Learning and Multi-Faceted Feature Engineering

Abstract

The increasing use of cloud-based technologies in education system will help a deeper understanding of adoption behaviors and predictive indicators. This study proposes an ensemble learning-based framework to predict the adoption of cloud-based educational platforms by leveraging multi-faceted feature engineering and educational data mining techniques. And different features have been extracted like behavioral, temporal, and content-specific features, video usage, file interactions and access time metadata all these are extracted from student activity logs and platform metadata. These features were used to train a stacked ensemble model stacked with LightGBM, XGBoost, and Random Forest classifiers. The proposed model achieved a MSE and RMSE 0.087 and 0.105, with a R2 error of 0.07, outperforming individual base models. In addition to performance metrics, explainable AI techniques such as SHAP (SHapley Additive exPlanations) were also applied to assess the model interpretability. SHAP summary plots and force plots revealed that features like ebook, files, and google_drive had the most substantial impact on model predictions, with SHAP values reaching up to ±0.05.