Subscriber Voice Call Metrics for Telecom Churn Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Telecommunication network operators face challenges in identifying and addressing individual subscriber issues that lead to poor network performance, which are often overlooked in conventional averaged metrics, resulting in subscriber churn.
Innovation Solution
A system and method that analyzes call detail records (CDRs) to identify negative voice call events for individual subscribers, determines subscriber-centric voice call metrics, and uses machine learning to predict churn risk, enabling targeted actions to improve communication quality and reduce churn.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional averaged performance metrics are used to evaluate network performance, then aggregate network performance can be assessed, but individual subscriber issues leading to poor service quality are overlooked
Solution Approach 1:
The patent segments the aggregate network performance metrics into individual subscriber-level metrics. Instead of calculating one averaged metric for the entire network, the system computes separate performance metrics for each subscriber based on their unique call patterns, dropped calls, blocked calls, and service experiences. This segmentation enables precise identification of individual service quality issues while maintaining manageable complexity through automated per-subscriber calculations.
Solution Approach 2:
The patent applies local quality by tailoring performance evaluation to each individual subscriber's specific service experience rather than applying a uniform aggregate metric. Each subscriber receives a customized performance assessment based on their unique call history, service patterns, and quality of service experiences, allowing the system to identify and address local service quality issues specific to each subscriber.
2Reliability
If subscriber-specific analysis is implemented to identify individual service issues, then service quality for individual subscribers can be improved, but computational resources and processing time increase
Solution Approach 1:
The patent implements preliminary action by pre-calculating and storing key performance indicators for each subscriber, such as counts of dropped calls, blocked calls, and service quality metrics over specific time periods. These pre-computed metrics are stored in data structures for efficient retrieval and analysis, eliminating the need for real-time computation of complex metrics when identifying service issues or predicting churn risk.
Solution Approach 2:
The patent substitutes complex real-time computational mechanisms with data retrieval and analysis of pre-computed metrics. Instead of performing intensive real-time calculations to assess service quality, the system retrieves stored performance data and uses machine learning models to predict churn risk, significantly reducing computational energy consumption while maintaining reliable service quality assessment.
3Quantity of substance
If conventional aggregate metrics are used, then overall network performance can be maintained, but subscriber churn risk increases due to unaddressed individual issues
Solution Approach 1:
The patent implements feedback by continuously monitoring individual subscriber service quality metrics and using this information to predict churn risk. The system provides feedback loops where service quality data is collected, analyzed through machine learning models, and used to identify subscribers at risk of churn, enabling targeted interventions to improve service quality and retain subscribers while maintaining overall network coverage.
Data Source
AI summary
A method of adapting a communication network to improve communication quality. The method comprises, for each of a plurality of subscribers, analyzing call detail records (CDRs) of the subscriber by an application executing on a computer system to identify negative voice call events; for each negative voice call event, associating a location of a subscriber communication device at the time of the negative voice call event by the application to the negative voice call event; for each of the plurality of subscribers, determining a subscriber-centric voice call metric for the subscriber by the application based on a count of negative voice call events of the subscriber for each of a plurality of one hour intervals; for each subscriber, determining an average subscriber-centric voice call metric by the application by averaging the subscriber-centric metric values determined for each of the plurality of one hour intervals; and taking action accordingly.