Network Device Classification with Hierarchical Confidence-Gated Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing number and diversity of network-connected devices pose challenges in effectively classifying entities for network security, leading to improper classification, resource inefficiency, and difficulty in applying appropriate security measures due to imbalanced labels, hierarchical classification issues, and resource-intensive machine learning approaches.
Innovation Solution
Utilizing multiple machine learning models organized in a hierarchical structure to perform classification at varying granularities, with each model focusing on specific levels of classification, and employing confidence thresholds to ensure accurate and efficient entity classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single machine learning model is used for entity classification, then the system is simple to implement, but classification accuracy deteriorates due to imbalanced labels and hierarchical classification challenges
Solution Approach 1:
The patent divides the classification task into multiple hierarchical levels, with each machine learning model responsible for a specific granularity level. This segmentation allows each model to focus on particular classes, improving accuracy for imbalanced labels while maintaining manageable system complexity through modular architecture.
2Measurement precision
If multiple machine learning models are used to improve classification accuracy, then classification precision improves, but resource consumption increases
Solution Approach 1:
By segmenting the classification task across hierarchical levels, the patent enables more efficient resource utilization. Each model processes only relevant features for its specific granularity, avoiding redundant computations that would occur in a monolithic model, thus reducing overall resource consumption while maintaining high accuracy.
3Use of energy by moving object
If traditional classification approaches are used, then resource consumption is low, but the system cannot effectively handle diverse and increasing number of network devices
Solution Approach 1:
The patent implements a dynamic hierarchical classification system where models can be selectively applied based on the entity type and required granularity. This dynamic approach allows the system to adapt to diverse device types efficiently, using more computational resources only when necessary for accurate classification of complex devices while conserving resources for simpler cases.
Data Source
AI summary
A system is configured to obtain data associated with a first device, based on network traffic. The system determines a first classification for the first device based on the data, including to determine that a first confidence level that is associated with the first classification satisfies a first threshold. In response to the first confidence level satisfying the first threshold, the system selects at least a last model from a plurality of machine learning models based on the first classification. The system determine a last classification for the first device based on the last model, including determining that a last confidence level that is associated with the last classification satisfies a last threshold. The system stores at least one of the first classification and the last classification.