Using Machine Learning for Cellular Network Data Diagnostics for Fault Detection & Prevention – Case Study by ThirdEye 2018-06-18T07:48:28+00:00
SUCCESS STORY

A Telecommunications Success Story –

Incorporating Machine Learning for Real-Time Network Data Diagnostics for Fault Detection and Prevention.

THE CUSTOMER

ThirdEye’s client is a Fortune 500 multinational Telco with a presence in over 100 countries. It provides research and innovation critical to the evolution of Global Telecommunications. From enabling telecommunication infrastructure and the Internet of Things to emerging applications in virtual reality and digital health, the client is shaping the future of technology and transforming the experience of the connected world.

BUSINESS GOALS

The hardware environment includes a wide variety of network equipment generating operations log data.
Interacting with each equipment and their associated dataset points and control schemes for accurately identifying and predicting problem areas using traditional engineering formulas was proving to be a challenge. Furthermore, with a multitude of possible equipment combinations & settings and a massive amount of user metadata, manually determining the optimal efficiency of the network was simply a daunting task.

The client needed a more efficient way to derive business value from network data.

The primary objectives being:

  • Automating the evaluation and prediction of problem areas and their impacts on network efficiencies.
  • Improving customer satisfaction, by proactively identifying and resolving problems faced by them.

THE SOLUTION

The technical aim of the project involved building a predictive model to find and resolve problem occurrences by analyzing data from a multitude of Base receiver stations (BTS), Radio Network controllers (RNC) and user equipment. Log data from the systems was collected and analyzed to predict problems in real time significantly alleviating the problem of manual log analysis.

The sequence of steps involved:

1. Identifying and categorizing sets problem from the reporting systems.
2. Analyzing time stamps associated with each problem occurrences and identifying and selecting appropriate
parameters that are valid and meaningful for that time instance.
3. Cluster analysis of the problem description and assigning a unique identifier for anomalies that occur
within the systems.
4. Processing data for training of the predictive models.
5. Validating the data and the predicting results to confirm the accuracy of the Engine, as developed.

Data Science & Machine Learning methodologies incorporated

Feature Reduction:
Feature Selection Algorithm: Max-Dependency, Max-Relevance, and Min-redundancy (mRMR) for pattern classification – This method was used to overcome the challenge defined as “curse of dimensionality” which is a well-known problem that predictive models don’t work on a large number of features

Using both flavors of Feature Selection algorithms, a hybrid approach as detailed below was incorporated.

  • Using a Filter approach to a get a relatively large set of features, agnostic to the predictive algorithm.
  • Applying a Wrapper approach of the features set from the previous step to arrive at a smaller set of features.
    This step was linked to the predictive algorithm that was developed by ThirdEye.

Other Predictive Algorithms Used:

  • Naïve Bayesian
  • Logistic Regression
  • Decision Tree

VALUE CREATED

The overall impact of this project on profitability was in the millions of dollars.
The following combination of factors helped in the op-ex reduction:

  • Fast, consistent & accurate fault localization
  • Ability to detect and respond to issues with limited manual interventions
  • Use of the system to identify and accelerate upgrades
  • Improved customer service

DATA ANSWERS

ThirdEye answers your data questions and offers actionable insights, real-world experiences and strategic recommendations.

Contact us