Improving Accuracy and Interpretability of Educational App Recommendations via Hybrid Machine Learning and Deep Learning Techniques

As the usage of mobile apps in digital education is growing, it has become more critical to provide accurate and personalised recommendation systems that can enhance user engagement and learning effectiveness. The difficulty of collaborative filtering techniques is caused by the sparseness, dynamism and time dependence of crowdsourcing educational information. As a means of solving these problems, a mobile app recommendation framework is created with the help of a crowdsourced educational feedback dataset from Kaggle. It was methodologically preprocessed using timestamp creation, discard of irrelevant attributes, label encoding, and feature engineering (pivoting) and 80/20 train-test split. A number of machine and deep learning models are used including SVD, SVD++, NMF, SlopeOne, variations of KNN, CoClustering, BaselineOnly, NormalPredictor, GRU, LSTM, Stacked Autoencoder, Gated Autoencoder and hybrid LSTM-GRU. Explainable variance, MAE and RMSE are used to assess the model performance. The experimental results show that the hybrid LSTM-GRU consistently gives the minimum mean RMSE value of 0.267332 and mean MAE value of 0.118370 with the largest individual performance at k = 16 (RMSE = 0.005066, MAE = 0.003821). LIME and SHAP enhance the explainability, and user interface with Flask enables practical applicability. This technique increases the accuracy, robustness and interpretability of the mobile educational app recommendation systems.