Privacy enhanced course recommendations through deep learning in Federated Learning environments

The increasing concerns around data security and privacy among users have significantly pushed the interest of the research community towards developing privacy-preserving recommendation systems. Amidst this backdrop, our study introduces a novel course recommendation methodology leveraging Federated Learning (FL) coupled with advanced Deep Learning techniques. This method executes the recommendation process across local nodes through several stages, including agglomerative matrix formulation, course clustering, bi-level matching, identification of learner-preferred courses, and ultimately, course recommendation. Notably, course clustering is achieved through Deep Fuzzy Clustering (DFC), while Deep Convolutional Neural Networks (DCNN) are employed for the recommendation phase. The efficacy of our DFC-DCNN-FL approach is rigorously evaluated based on several metrics: accuracy, False Positive Rate (FPR), loss function, Mean Square Error (MSE), Root MSE (RMSE), and Mean Average Precision (MAP). The results demonstrate remarkable performance with scores of 0.909, 0.116, 0.126, 0.291, 0.539, and 0.925, respectively.