Skip to main content

Requirements

Here is presented the requirements elicitation methods, functional requirements (divided by the respective components) and non-functional requirements.

Requirements Elicitation​

This process was divided into four parts. The first part consisted of investigating the project's theme and its context to obtain maximum knowledge in the field of telecommunications and 5G networks. Next, we had to evaluate the most correct way to relate this to the data engineering and Machine Learning part of the project. After that, we explored the state of the art, evaluating other research works in the area and gathering important information for discussion among us.

Functional Requirements​

Data Collection​

The system:

  • must collect and store raw data in a time series data base.
  • system should support JSON and CSV data formats.
  • system should perform data pre-processing, including cleaning, normalization, and aggregation to ensure quality and consistency.

Model​

The system should:

  • train machine learning (ML) models using frameworks such as PyTorch or Scikit-learn.
  • allow model training with stored data.
  • implement automation in model training, supporting continuous re-training based on new data or data drift.
  • validate and test the models using the obtained metrics.
  • support a continuous deployment mechanism to automatically move validated models from the test environment to production.
  • be able to identify the relevant features for anomaly prediction.
  • not allow the use of future data to train the model.

Monitoring Dashboard​

  • Must show anomaly alerts.
  • Must show relevant metrics.

External Integration​

The system should:

  • provide APIs so that external functions can subscribe to anomaly and failure events.
  • enable compliance with 5G architecture standards.

Non-Functional Requirements​

Scalability​

  • The system must efficiently process large volumes of data with a maximum processing latency of 1 milliseconds.

Performance​

  • To support real-time analytics, data processing should have minimal latency and a response time inferior to 1 millisecond.
  • ML inference APIs should provide responses within 1 millisecond for fast decision-making.

Security​

  • The system must be GDPR compliant and keep all data on-premise.

Maintainability​

The system:

  • should use modular components to allow easy updates and debugging.
  • must allow modules to be replaced by others with higher performance, with minimal impact on other modules.
  • must be easily adaptable for deployment in several network environments.
  • must follow good MLOps practices, guaranteeing modularity, reproducibility and complete automation of the ML lifecycle.

Interoperability​

  • The system must be interoperable, making available APIs and adopting machine learning frameworks.