| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 1.53 MB | Adobe PDF |
Authors
Abstract(s)
In order to minimize the impact of potential failures of assets within its telecommunications
infrastructure in Portugal, NOS is interested in improving the efficiency of current decision processes which are led by human operators. Currently, an expert team manually selects response
teams to deal with occurred failures, based on a pre-compiled set of policies to classify alarms.
However, NOS aims to leverage ML-based techniques to help the expert team select the most
appropriate response team to deal with an occurred technical issue.
Towards that goal, one of the main challenges lies in the highly dynamic nature of the alarms
issued by the company’s large and diverse infrastructure. In this project, we explore the design of
a fully automated failure analyzer based on neural networks, called Autodispatcher, to tackle this
challenge.
The Autodispatcher framework will function as a decision tree, capable of efficiently analyzing
and processing real-time alarms emitted by the infrastructure monitoring tools. It is designed to
have four stages, including two stages which are the main focus of this work: a string frequency
analysis method and a neural network approach for word analysis.
To ensure the efficiency and effectiveness of the Autodispatcher, regular model updates and
retraining are conducted. This allows the system to adapt to changes in the infrastructure and
incorporate new data patterns that may arise over time. Additionally, human operators play a
considerable role in supervising and validating the Autodispatcher’s decisions, because they have
the ability to review and adjust the system’s recommendations.
The main goal of the Autodispatcher is to minimize response time and optimize resource allocation. By automating the process of selecting response teams, NOS can significantly improve the
efficiency of its operations. This reduces the impact of failures on the infrastructure and enhances
the overall customer experience by ensuring timely resolutions to technical issues.
Description
Trabalho de projeto de mestrado, Ciência de Dados , 2023, Universidade de Lisboa, Faculdade de Ciências
Keywords
Machine Learning Redes Neuronais Árvores de Decisão Python TensorFlow Trabalhos de projeto de mestrado - 2024