Network KPI Forecasting via Heterogeneous Data Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network assurance systems face challenges in forecasting key performance indicators (KPIs) due to the heterogeneity of network entities and data, which is often partially structured, numerical, and irregularly sampled, making it difficult to predict network issues like tunnel failures in a network-specific manner.
Innovation Solution
A service that receives input data from networking entities, including synchronous time series, asynchronous events, and entity graphs, clusters entities by type, selects relevant machine learning model features, and trains models to forecast KPIs for specific entity clusters, using a combination of synchronous and asynchronous data to build scalable and adaptable forecasting models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are trained on heterogeneous network data from multiple entity types, then forecasting accuracy for network-specific KPIs is improved, but data processing complexity and model training difficulty increase
Solution Approach 1:
The patent segments the heterogeneous network data into distinct types (synchronous time series data, asynchronous event data, entity graph data) and processes each type through dedicated pipelines before feeding into the machine learning model. This segmentation reduces the complexity of handling diverse data formats by treating them separately through standardized transformation processes.
Solution Approach 2:
The patent introduces an intermediary data processing layer that transforms heterogeneous network data from various entity types into a unified representation suitable for machine learning consumption. This intermediary layer handles the complexity of data normalization, feature extraction, and integration, shielding the downstream model from data heterogeneity issues.
2Measurement precision
If network data from multiple entity types is collected and processed, then the forecasting model becomes more comprehensive and accurate, but the time required for data collection and processing increases
Solution Approach 1:
The patent implements preliminary data collection and processing actions by continuously gathering synchronous time series data, asynchronous event data, and entity graph data from network entities before forecasting is needed. This pre-processing and storage of data in optimized formats enables rapid query and model training without time-consuming data collection during the forecasting operation.
Solution Approach 2:
The patent establishes continuous data collection and processing operations that run alongside network operations, maintaining a persistent dataset ready for forecasting. This continuous operation eliminates interruptions and delays by keeping data flowing through the processing pipeline without idle time between data collection and model consumption.
3Adaptability or versatility
If the system handles heterogeneous and irregularly sampled network data, then the forecasting model becomes more adaptable to different networks, but the complexity of data normalization and feature selection increases
Solution Approach 1:
The patent applies parameter changes by dynamically adjusting data normalization parameters and feature selection criteria based on the specific characteristics of each network entity type. The system modifies transformation parameters (such as scaling factors, binning thresholds, and feature weights) to adapt to the unique data distributions and sampling patterns of different network entities, enabling versatile forecasting across heterogeneous networks.
Data Source
AI summary
In one embodiment, a service receives input data from networking entities in a network. The input data comprises synchronous time series data, asynchronous event data, and an entity graph that that indicates relationships between the networking entities in the network. The service clusters the networking entities by type in a plurality of networking entity clusters. The service selects, based on a combination of the received input data, machine learning model data features. The service trains, using the selected machine learning model data features, a machine learning model to forecast a key performance indicator (KPI) for a particular one of the networking entity clusters.