Enhancing QoS with Deep Learning:

A Comprehensive Literature Review on Model Optimization and Advanced Data Labeling

Abstract

This literature review is based on 70 research studies and deals with the issue of optimizing deep learning algorithms and data labeling for network Quality of Service (QoS) classification. It was found that 60% (42 out of 70) of the studies performed showed an inherited optimization need in leveraging deep learning techniques and 67.14% (47 studies) showed the same in data labeling techniques. The intricacy involved in computational processes pertaining to deep learning models poses a significant challenge, as it entails considerable resource investment during the phases of model training and execution. Models can be optimized through hyperparameter tuning, changing network architecture or adding other strategies such as transfer learning which improve reliability and scalability. Also, LSTM networks and other related techniques are effective in capturing temporal phenomena surrounding network traffic, thus enhancing the model's relevance to real-world situations. Equally important is the optimization of data labeling, where challenges such as class imbalance can be resolved through oversampling, undersampling or generating synthetic data. The inclusion of accurate, complete and consistent datasets improves training efficiency. The author’s conclusions accentuate that both enhancing the deep learning process and data labeling techniques should be considered for the effective and accurate development of network QoS classification.