4G-Vulnerability-predictor-Azure-MLops

This project showcase the 4G vulnerability prediction, where training and deployment is configured over Azure MLops Platform.

Use-Case:

Anamoly detection in 4G cellular networks.

• It is a classficiation dataset where ML algorithms are used to detect abnormal network behaviour based on certain features. Network Optimization team aims to train ML system to classify current acivity as normal(0) or unusual(1).
• Normal behavior implies no resource re-configuration, while unusual behavior require base station adjustments.
• Dataset sourced from a real LTE deployment over two weeks, gathering metrics from 10 base-station every 15 minutes.
• Data provided in a CSV file, each row representing a sample from a specified cell at a particular time.

Certain Data statistics:

• No of variables: 14.
• No of observations: 36904.
• Missing Cells: 0.
• Duplicate rows: 33.
• Total size in memory: 3.9 MB.

Data Snapshot:

Code_Structure: Information about main code files and folders.

• /Data: Contains the data in CSV format.
• /training: Includes the training code and related configuration files to be executed over Azure ML.
• /deployment: Contains the code to train the best model, download and deploy it to create the end-point.

Steps defined under Azure_Devops:

There are two pipelines defined here:

• Azure_Training_Build is the pipeline to train the machine learning model.
• Deployment pipeline is the pipeline to deploy the trained model and produce the end point. The steps are defined in the image below. Azure_Training_Build: This is defined as Steps within Azure devops as shown below:

Deployment_Pipeline: Deployment pipeline is defined as a pipeline.yaml under deployment/pipeline.yaml file.

Relevant Screenshots:

• Mlflow generated experiments: Experiment with best accuracy:
• Best model is saved as onnx file.