QoE-Driven IoT Network Optimization using CNN,Transformer Attention and Federated Learning

Abstract

Quality of Experience (QoE) has also become a user-centric performance metric in Internet of Things (IoT) networks, that is considered to be more realistic in reflecting user satisfaction than traditional Quality of Service (QoS) metrics. The current paper introduces a new hybrid model consisting of Convolutional Neural Networks (CNNs), Transformer-based attention models, and Federated Learning (FL) to predict real time QoE. The use of self attention improves the relevance of global features, while FL ensures data privacy and scalability between heterogeneous IoT clients. Experiments conducted in CICIDS 2018 and IoT-TON datasets demonstrate the model superior performance in accuracy (up to 98.63\%), F1-score, and cross-entropy loss compared to existing approaches. The proposed architecture is validated using QoS-to-QoE mappings and class-wise classification metrics, and is further benchmarked against recent deep and federated models. The results confirm that the framework is interpretable, privacy-preserving, and effective for QoE monitoring in large scale IoT environments.