Enhancing Mobile App Recommendations with Crowd source Educational Data Using Machine Learning and Deep Learning

Abstract

The increasing adoption of mobile applications in digital education has intensified the need for accurate and personalized recommendation systems to improve user engagement and learning effectiveness. Traditional collaborative filtering techniques face limitations when handling sparse, evolving, and time-dependent crowdsourced educational data. To overcome these challenges, a mobile app recommendation framework is developed using a crowdsourced educational feedback dataset obtained from Kaggle. The dataset is systematically preprocessed through timestamp generation, removal of irrelevant attributes, label encoding, pivot-based feature engineering, and an 80/20 train–test split. Multiple machine learning and deep learning models are implemented, including SVD, SVD++, NMF, SlopeOne, KNN variants, CoClustering, BaselineOnly, NormalPredictor, GRU, LSTM, Stacked Autoencoder, Gated Autoencoder, and a hybrid LSTM–GRU architecture. Model performance is evaluated using explained variance, mean absolute error (MAE), and root mean square error (RMSE). Experimental results demonstrate that the hybrid LSTM–GRU model consistently outperforms all baseline and standalone deep learning models, achieving the lowest mean RMSE of 0.267332 and mean MAE of 0.118370, with the best individual performance at k = 16 (RMSE = 0.005066, MAE = 0.003821). Explainability is further enhanced using LIME and SHAP, and practical applicability is supported through a Flask-based user interface. This approach significantly improves accuracy, robustness, and interpretability in mobile educational app recommendation systems.