Publication: Adaptación y calibrado de algoritmos de predicción para la identificación de ataques DDoS en redes de quinta generación
Loading...
Files
Official URL
Full text at PDC
Publication Date
2018
Advisors (or tutors)
Maestre Vidal, Jorge
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citations
Exportar
Abstract
El avance de las redes de telefonía móvil hacia su quinta generación, popularmente conocida como 5G, viene de la mano de una colección de tecnologías emergentes que brinda importantes mejoras en sus principales indicadores de desempeño, como su rendimiento, eficiencia, ahorro energético o movilidad. También permiten desarrollar capacidades de autoorganización basadas en el estudio de observaciones en el entorno de monitorización, dando un enfoque cognitivo y holístico a sus mecanismos de respuesta a incidencias. Con el fin de contribuir a su desarrollo, el trabajo realizado se centra en la anticipación de eventos en red, habiéndose desarrollado una estrategia de predicción adaptativa que tiene en cuenta la gran heterogeneidad de fuentes de información y la no estacionariedad, inherentes a los escenarios de red venideros. Esto se ha logrado mediante la implementación de estrategias de aprendizaje automático para la selección de los mejores algoritmos según el contexto, y la evolución de su calibrado acorde a las variaciones de las observaciones. El método propuesto ha sido evaluado a partir del estándar de evaluación funcional M3-Competition y en un caso de uso específico: la detección de ataques de denegación de servicio distribuidos. Para esto último se ha recopilado una colección de muestras de tráfico de red capturados en dispositivos de diferente naturaleza, a partir de las cuales se han extraído y analizado indicadores propios de este tipo de amenazas. La amplia experimentación realizada ha arrojado resultados muy prometedores, indicando interesantes líneas de trabajo futuro.
The advances of wireless mobile networking towards its fifth generation, popularly known as 5G, arrive hand in hand with a collection of emerging technologies that provide important improvements in terms of key performance indicators related among others, with effectiveness, efficiency, energy consumption and mobility. They also facilitate the development of self-organization capacities based on studying observations on the monitoring environment, thus bringing cognitive and holistic solutions to their incident response mechanisms. In order to contribute to their development, the performed work focuses on the problem of anticipating network events. With this purpose, a novel adaptive prediction strategy that takes into account the great heterogeneity of data sources and the non-stationarity inherent in the forthcoming landscape, has been developed. This has been achieved through the implementation of machine learning methods for selecting the best algorithms according to the context, and by making evolve their calibration based on variations at traffic observations. The proposed approach has been evaluated in the grounds of the functional evaluation standard M3-Competition. In addition, it was deployed on a specific use case: the detection of distributed denial of service attacks. For the latter, a collection of network traffic samples captured from devices of different nature has been gathered, from which classical indicators of this kind of threats have been extracted and analyzed. It is worth to highlight that the extensive experimentation displayed very promising results, thus pointing out interesting lines of future research.
The advances of wireless mobile networking towards its fifth generation, popularly known as 5G, arrive hand in hand with a collection of emerging technologies that provide important improvements in terms of key performance indicators related among others, with effectiveness, efficiency, energy consumption and mobility. They also facilitate the development of self-organization capacities based on studying observations on the monitoring environment, thus bringing cognitive and holistic solutions to their incident response mechanisms. In order to contribute to their development, the performed work focuses on the problem of anticipating network events. With this purpose, a novel adaptive prediction strategy that takes into account the great heterogeneity of data sources and the non-stationarity inherent in the forthcoming landscape, has been developed. This has been achieved through the implementation of machine learning methods for selecting the best algorithms according to the context, and by making evolve their calibration based on variations at traffic observations. The proposed approach has been evaluated in the grounds of the functional evaluation standard M3-Competition. In addition, it was deployed on a specific use case: the detection of distributed denial of service attacks. For the latter, a collection of network traffic samples captured from devices of different nature has been gathered, from which classical indicators of this kind of threats have been extracted and analyzed. It is worth to highlight that the extensive experimentation displayed very promising results, thus pointing out interesting lines of future research.
Description
Trabajo de Fin de Grado en Ingeniería Informática. Facultad de Informática, departamento de Ingenierá del Software e Inteligencia Artificial, curso 2017/2018