<![CDATA[The rapid increase in interconnected devices, commonly known as the Internet of Things (IoT), has significantly impacted various sectors, enhancing services in energy, transport, health, and more. Unfortunately, that means consumers are facing increasing challenges in choice of quality IoT devices and services alike. Importantly, traditional methods of recommendation are largely reliant on collaborative or content-based filtering, which suffer from problems such as sparsity of data and the cold start problem that all result in potentially large inaccuracies. In this regard, this study supplies a unique QoS prediction approach with Generative Adversarial Network (GAN) and Gated Recurrent Unit (GRU), i.e., GRU-GAN is proposed to address these challenges. This approach maps the QoS matrix on service call records, involving user attributes and historical QoS records as a time series to train GRU-GAN model. In the GAN, the generator is trained to predict realistic QoS values and then discriminator evaluates classifying these predictions. We experimentally show the efficiency of our model. Our GRU-GAN model consistently outperforms traditional QoS prediction methods showing lower RMSE and MAE with regards to different data densities. More concretely, it had an RMSE of 0.16 and MAE of.05 with data density at 5%, and performed best across all the model as your increased data availability beyond that scale. In conclusion, the GRU-GAN model offers a robust solution for QoS prediction in IoT ser- vice recommendations, effectively handling data sparsity and enhancing prediction accuracy.]]>
