Cellular Network Traffic Clustering for ML Model Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning models for cellular network traffic management are inefficient in adapting to diverse network topologies and traffic conditions, requiring significant time and effort to build and update, and often result in complex models with reduced accuracy.
Innovation Solution
The approach involves clustering cellular network measurement data based on similarities in performance indicators and traffic load, assigning machine learning models to each cluster, and iteratively refining the clustering and model training until an acceptable error rate is achieved, allowing for deployment of models that balance accuracy and simplicity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are deployed to individual network nodes, then model accuracy is improved, but device complexity and time/effort for building and tuning models increases
Solution Approach 1:
The patent merges multiple individual machine learning models into a single centralized model deployed at a control hub. This consolidation reduces the number of models from many individual node models to one unified model, directly addressing the complexity issue while maintaining predictive capabilities through centralized data aggregation from all network nodes.
Solution Approach 2:
The centralized machine learning model serves multiple network nodes simultaneously, making it a universal solution. The single model handles traffic prediction for numerous nodes by processing aggregated data from all of them, eliminating the need for separate specialized models at each node while maintaining comprehensive coverage.
2Measurement precision
If machine learning models are deployed to individual network nodes, then model accuracy is improved, but time and effort for building and tuning models increases
Solution Approach 1:
The patent combines the model building and tuning processes into a single centralized effort rather than requiring separate development cycles for each individual node model. This unified approach reduces the total time and effort required by eliminating redundant work across multiple nodes.
Solution Approach 2:
The centralized model is trained in advance using aggregated historical data from all network nodes before deployment. This preliminary training action allows the model to be ready for immediate use across multiple nodes without requiring subsequent tuning at each individual node, significantly reducing implementation time.
3Device complexity
If a single machine learning model is used for the entire area, then device complexity is reduced, but model accuracy deteriorates
Solution Approach 1:
The patent segments the network data by geographic areas or regions before feeding it to the centralized model. This segmentation allows the single model to process distinct regional patterns separately, maintaining accuracy by preserving local characteristics while still achieving simplicity through centralized deployment and management.
Data Source
AI summary
The described technology is generally directed towards data clustering for network traffic modeling. Cellular network measurement data from different geographic areas can be separated into clusters based on similarities in network performance indicators, cell traffic load data, their changing pattern over time, and/or other metrics. A machine learning model can then be assigned to each cluster, and the machine learning models can be trained to make network traffic control decisions under conditions exhibited in their respective clusters. If the error rate of the trained machine learning models is acceptable, then the machine learning models can be deployed for use at network equipment. If the overall error rate is not acceptable, then the cellular network measurement data can be re-separated into a larger number of clusters, and machine learning models can again be trained for each cluster. The re-separation of data and re-training of machine learning models can repeat until the error rate is acceptable and the machine learning models can be deployed.