Network Assurance Service Predicting Roaming Onboarding Delays
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current network assurance systems face challenges in predicting and forecasting roaming issues in wireless networks due to the complexity of network dynamics and the large number of parameters involved, which makes it difficult to accurately assess and prevent onboarding delays and user experience impacts.
Innovation Solution
A network assurance service applies labels to feature vectors of network characteristics associated with wireless access points, trains machine learning-based classifiers to predict onboarding delays, and selects the most performing classifier to monitor network characteristics, using a hybrid classification framework to address the imbalanced labeled data and radio-specific signal variations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning classifiers are trained to predict onboarding delays in wireless networks, then prediction accuracy for roaming issues is improved, but the complexity of the network assurance system increases
Solution Approach 1:
The patent segments the network assurance system into multiple independent machine learning classifiers, each specialized for predicting specific types of onboarding delays (e.g., association delays, authentication delays). This segmentation allows the system to achieve high prediction accuracy for different delay types while managing overall system complexity through modular architecture.
Solution Approach 2:
The patent employs parameter changes by training classifiers with different feature sets and hyperparameters optimized for specific delay types. Each classifier uses tailored parameters such as feature selection, training data windows, and model complexity levels, enabling accurate predictions without uniformly increasing the entire system's complexity.
2Reliability
If multiple machine learning classifiers are trained to handle different types of onboarding delays, then the ability to predict specific roaming issues is improved, but the training time and computational resources increase
Solution Approach 1:
The patent implements preliminary action by pre-training multiple specialized classifiers during system setup and deployment phases. The classifiers are trained in advance on historical network data and stored for immediate use during operation, eliminating the need for real-time training and reducing operational training time to zero.
Solution Approach 2:
The patent applies dynamics by implementing adaptive classifier selection that dynamically chooses which pre-trained classifier to use based on current network conditions and the specific type of onboarding delay being predicted. This dynamic approach optimizes computational resource usage by activating only the necessary classifiers rather than running all classifiers continuously.
3Measurement precision
If a hybrid classification framework is used to address imbalanced labeled data, then prediction accuracy for rare onboarding delays is improved, but the complexity of the training process increases
Solution Approach 1:
The patent applies local quality by implementing different training strategies and data sampling methods for different classes of onboarding delays. For rare events, the system uses oversampling or specialized loss functions locally applied to those specific classes, while common events use standard training approaches. This localized approach improves rare event detection without uniformly complicating the entire training process.
Data Source
AI summary
In one embodiment, a network assurance service applies labels to feature vectors of network characteristics associated with a plurality of wireless access points in the network. An applied label for a feature vector indicates whether the access point associated with the feature vector experienced a threshold number of onboarding delays within a given time window. The service, based on the feature vectors and labels, trains a plurality of machine learning-based classifiers to predict onboarding delays, and uses one or more of the trained plurality of classifiers to predict onboarding delays for a particular access point. The service calculates one or more classifier performance metrics for the one or more classifiers based on the predicted onboarding delays for the particular access point. The service selects a particular one of the classifiers to monitor the network characteristics associated with the particular access point, based on the one or more classifier performance metrics.