Digital Developer Conference on Data and AI: Essential data science, machine learning, and AI skills and certification Register for free

Predicting telecom call drops with AI

In the telecom domain, a call drop is a situation where a call on a wireless network is disconnected before the caller ends the call. These call drops can happen for several reasons. Inadequate coverage such as lack of tower infrastructure, improper network planning, and non-optimization of networks can greatly affect the stability of the network. Sometimes, the network capacity is simply not being ramped up at the same pace as usage, which results in overloaded networks.

Common everyday situations can also affect network coverage. Multi-storied buildings within a city can cause subscribers in surrounding buildings to lose cell reception. And, switching between towers such as when someone is traveling or moving while on the phone can increase the number of dropped calls whenever the call is transferred from one base transceiver station (BTS) to another, especially in the case of overloaded networks.

This solution demonstrates a telecom call drop scenario that uses the following IBM products:

  • IBM Cloud Pak® for Data
  • IBM Watson® OpenScale™
  • Cognos® Analytics.

Within the solution, data can come from multiple database sources, for example, an internal Db2® Warehouse within the IBM Cloud Pak for Data instance, or other external sources like IBM Db2 on Cloud, Netezza Performance Server, an Oracle database, or a Postgres database.

Data virtualization integrates them all into one database source. Using a built-in notebook service, a time-series model predicts call drops in the next 24 hours, and there is a call-drop prediction model for each cell tower. These models are monitored for quality and fairness using Watson OpenScale. A Cognos Analytics dashboard shows an overall region view of the call drop scenarios. With the help of Watson OpenScale, the time-series model is output in a graph, along with the model’s performance improvements.

This use case provides an end-to-end solution, starting with:

  • Collecting and aggregating data
  • Building and monitoring machine learning models used to predict call drops for a given cell tower
  • Creating, managing, and deploying a dashboard to gain insights about the built machine learning system

Building this solution consists:

  • Querying across distributed data sources as one
  • Monitoring your machine learning models
  • Building dashboards in Cognos Analytics on IBM Cloud Pak for Data
  • Predicting, managing, and monitoring the call drops of cell towers using IBM Cloud Pak for Data

Query across distributed data sources as one: Data virtualization for data analytics

Data can reside in multiple data marts. And data virtualization, within IBM Cloud Pak for Data, can perform data integration seamlessly. Data virtualization can connect to data wherever it resides and provides the ability to view, access, manipulate, and analyze data without the need to know or understand its physical format and location. Data virtualization creates virtual tables to join data from different data sources, then allows running queries against the resulting virtual table.

Learn how virtualization can be achieved on various data sources with an Oracle database hosted on Amazon Web Services with IBM Cloud Db2 Warehouse and IBM Db2 on Cloud. You can also connect to Netezza Performance Server or other databases by selecting from the menu.

Monitor your machine learning models

Today, businesses are increasingly certain that AI will be a driving force in the evolution of their industries over the next few years. To successfully infuse AI into your product or solution, there are many factors that challenge its widespread adoption. These might include:

  • Building trust
  • Algorithm bias
  • Making decisions explainable

Monitor machine learning models flow diagram

Learn how to build an app to monitor and deploy open source machine learning models using Watson OpenScale on IBM Cloud Pak for Data or IBM Cloud to configure fairness, quality, and explainability monitors for each cell tower’s model.

Build dashboards in Cognos Analytics on IBM Cloud Pak for Data

Suppose you’re a telecom servicer provider responsible for determining why call drops are happening frequently for any selected towers. You must understand the reasons for the call drops along with the contributing factors that influence the call drops for the selected tower. You get these insights from the Python model output, and using these details and the IBM Cognos Analytics dashboard feature, you should be able to build a dashboard that depicts call-drop prediction for the next 24 hours.

Learn how you can create a dashboard using IBM Cognos Analytics on IBM Cloud Pak for Data as well as how to import and export the dashboard binaries in the IBM Cloud Pak for Data environment.

Predict, manage, and monitor the call drops of cell towers using IBM Cloud Pak for Data

The final part of this solution is using a model to predict call drops. This model is trained on the various call drop failure situations. With the help of an interactive dashboard, a time-series model helps to provide a better understanding of the call drops. A Cognos Analytics dashboard provides an overall region view of the call drop scenarios. And with the help of Watson OpenScale, the time-series model is output in a graph, along with the models performance improvements.

Predict, manage, monitor call drops flow diagram