Enhancing Mobile App Recommendations With Crowdsourced Educational Data Using Machine Learning and Deep Learning

In the rapidly evolving digital landscape, personalized recommendations have become essential for enhancing user experience. Machine learning models analyze user behavior patterns to suggest relevant entertainment, education, or e-commerce content. Mobile devices make it easier to gather educational data through crowdsourcing, which opens new possibilities for improving app recommendation algorithms. This paper provides valuable methodologies for scalable student recommendation and educational systems, highlighting DL's advantages over CF in handling sparse, time-sensitive datasets. The objective of this study is to recommend apps to university students by category based on app usage patterns. Data was used to evaluate these 806 university students to train the Collaborative Filtering (CF) and Contemporary Deep Learning (DL) models. The results demonstrate that Gated Recurrent Units (GRU) are the best option for real-time, customized suggestions because of their capacity to simulate successive interactions and adjust to changing user behavior. The GRU yields the lowest mean errors MAE of 0.2246, RMSE=0.2516, and superior short-term predictions k=4 MAE of 0.1319 and RMSE of 0.1319. Other techniques, i.e., Stacked Auto-Encoder, exhibit the sign of overfitting with an MAE of 0.0001, whereas the LSTM and Graph Auto-Encoder perform below GRU with an MAE of 0.3453 and 0.8992. Although the CF techniques suffer from temporal dynamics and data sparsity, even the KNNBasic stands out among all CF algorithms with the lowest MAE of 0.548 and RMSE of 0.754, demonstrating the highest predictive accuracy.