Guarino, Idio (2024) Classification and Prediction of Communication and Collaboration Mobile Apps’ Traffic via Deep Learning Approaches: From Data Collection to Models’ Explainability. [Tesi di dottorato]
Anteprima |
Testo
Guarino_Idio_36.pdf Download (14MB) | Anteprima |
| Tipologia del documento: | Tesi di dottorato |
|---|---|
| Lingua: | English |
| Titolo: | Classification and Prediction of Communication and Collaboration Mobile Apps’ Traffic via Deep Learning Approaches: From Data Collection to Models’ Explainability |
| Autori: | Autore Email Guarino, Idio idio.guarino@unina.it |
| Data: | 10 Gennaio 2024 |
| Numero di pagine: | 192 |
| Istituzione: | Università degli Studi di Napoli Federico II |
| Dipartimento: | Ingegneria Elettrica e delle Tecnologie dell'Informazione |
| Dottorato: | Information technology and electrical engineering |
| Ciclo di dottorato: | 36 |
| Coordinatore del Corso di dottorato: | nome email Russo, Stefano stefano.russo@unina.it |
| Tutor: | nome email Pescapè, Antonio [non definito] |
| Data: | 10 Gennaio 2024 |
| Numero di pagine: | 192 |
| Parole chiave: | communication-and-collaboration-apps, encrypted traffic, deep learning, traffic classification, traffic prediction, XAI. |
| Settori scientifico-disciplinari del MIUR: | Area 09 - Ingegneria industriale e dell'informazione > ING-INF/05 - Sistemi di elaborazione delle informazioni |
| Depositato il: | 11 Gen 2024 07:58 |
| Ultima modifica: | 12 Ago 2026 05:36 |
| URI: | https://www.fedoa.unina.it/id/eprint/15590 |
Abstract
Internet traffic is constantly changing as networks and applications evolve. The COVID pandemic has greatly increased the use of communication and collaboration mobile apps (CC apps), significantly affecting the volume of traffic they generate. This requires appropriate management of network resources, also based on the multiple activities that can be performed by such apps. This Thesis focuses on the traffic generated by CC apps to understand their peculiarities and design advanced tools for their classification and prediction. Integrating these tools can result in a more flexible and adaptable network management strategy. To address modern traffic challenges, they use Deep Learning (DL) techniques that can adapt to frequently changing input data and improve performance on complex tasks. Given the centrality of data, experimental analyses exploit real traffic from popular CC apps, purposely collected and publicly released. First, a characterization of the traffic is provided to highlight its distinctive features. Next, a novel approach to traffic classification is designed to overcome experimentally verified state-of-the-art gaps, particularly by exploiting contextual representations of observed traffic. Innovative approaches for traffic prediction at both packet- and aggregate-level are then designed by evaluating the trade-off between performance and complexity. Finally, to address the limitations in transparency and reliability due to the black-box nature of DL models, the interpretability of their results is analyzed via eXplainable Artificial Intelligence techniques. In summary, this Thesis investigates and tackles the peculiar challenges of CC apps’ traffic from multiple perspectives. It provides new solutions to support classification and prediction and focuses on interpretability and reliability issues arising from DL adoption.
Downloads
Downloads per month over past year
Actions (login required)
 |
Modifica documento |
