Following our exploration of generative AI in the first part of this series, we now turn our attention to another AI technique in this second article. This article considers how discriminative AI models also have the potential to play a crucial role in the telecommunications sector, especially via OSS/BSS solutions

The objective of discriminative models

Discriminative models find the boundary between categories / classes within data sets to enable any new data points to be quickly classified / or categorised. The objective of discriminative models is to sort data into categories, in a similar mode to rock band Van Halen insisting on having a bowl of M&M candies backstage with all of the brown ones removed

Generative AI techniques calculate a probability score based on prior and likely probabilities, which allows it to generate new data points. Conversely, discriminative AI techniques calculate the boundary conditions such that data can be ruled in or out in a binary manner. Red, orange or green M&Ms are ruled in. Brown M&Ms are ruled out. This doesn’t allow for new data to be created, only supplied data to be sorted.

Discriminative AI relies on a training phase where each data point is associated with classes / labels. Then feature extraction is performed to identify features / boundaries that allow separation (e.g. colour of an M&M). The model is then trained to identify the boundary that separates classes (e.g. brown vs other colours). After training, the model is evaluated against a new data set (i.e. not the same as the training data set) to determine accuracy of the model and readiness to be applied to future data sets.

Using generative and discriminative AI models for network fault detection

In the context of OSS/BSS, discriminative AI models play a critical role in classification and decision-making tasks including:

  • Customer Churn Prediction: Utilising discriminative algorithms, telcos can analyse customer behaviour and predict potential churn, enabling targeted interventions
  • Network Fault Detection: OSS can leverage discriminative models to identify and predict network faults through comprehensive analysis of network telemetry data
  • Spam Call Detection: Discriminative models assist in classifying calls by the likelihood of being spam. This enables the carrier to notify users that an incoming call is potentially a spam caller, thus improving user experience
  • Similarity Search on Log and Telemetry Data: Anomaly detection and pattern recognition in unstructured log data can offer valuable insights into network performance and security. They also perform well for classification / segmentation of telemetry data (but don’t provide the additional insights of underlying distributions in the data like generative models do)
  • Fault Prediction: Identifying network components that are most likely to fail soon, based on historical data and real-time metrics
  • Traffic Management and Resource / Capacity Allocation: Analysing network traffic / usage patterns to identify areas of congestion helps operators (or self-optimisation algorithms) to identify areas to allocate resources more efficiently
  • Customer Segmentation: Classifying customers into different groups based on usage patterns, enabling personalised marketing, service offerings and user experiences
  • Quality of Service (QoS) Monitoring: Dynamically classify network transactions based on their QoS metrics like latency, jitter, and packet loss
  • Social / User Network Analysis: Classifying social connections to identify influencers or central activity nodes within a telecommunications network, which could be useful for capacity planning, marketing, information dissemination and other use-cases

Separating signal from noise: The rise of discriminative AI in network operations

These techniques help to separate the signal from the noise, allowing:

  • Enhanced Decision Making: By focusing on class boundaries, discriminative models provide accurate classifications, empowering decision-makers with highly targeted, data-driven insights
  • Resource Optimisation: Automating complex tasks like anomaly detection, usage patterns and customer segmentation strengthens the ability to allocate resources efficiently
  • Resilience, Compliance and Security: In the ever-changing landscape of network complexity, regulations and threats, these models offer rapid insights at speed and scale to ensure optimised operations


Discriminative AI offers exciting capabilities for the way OSS/BSS solutions operate, adding efficiency, accuracy and intelligence to various functions for network operators. If you haven't read the first part of this series focusing on generative AI in telco, you can find it [here]. Together, these two articles shed light on the complementary nature of discriminative and generative AI and their pivotal role in shaping the future of telecommunications.

By understanding both, telecommunications providers can craft more effective strategies, maximise operational efficiency / effectiveness and drive innovation in an increasingly competitive and fast-changing landscape.