ML-Based Anomaly Detection for Mobile RAN Root Cause Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for anomaly detection and root cause analysis in mobile networks are inadequate, as they rely on manual threshold settings, are prone to errors, and fail to detect contextual anomalies, leading to high false alarm rates and inefficiencies.
Innovation Solution
A system utilizing machine learning methods for automated anomaly detection and root cause analysis, which includes a network analysis platform, machine learning engine, anomaly detector, classification engine, and alert engine to identify and classify anomalies, and provide real-time alerts and recommendations for improving network performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If manual threshold-based anomaly detection is used, then the system is simple to implement, but it produces high false alarm rates and cannot detect contextual anomalies
Solution Approach 1:
The patent replaces manual threshold-based detection with machine learning models that automatically learn optimal thresholds from historical data. The system uses supervised learning algorithms to train classifiers on labeled anomaly data, substituting mechanical threshold-setting with intelligent automated detection that adapts to contextual patterns in network behavior.
Solution Approach 2:
The system dynamically adjusts detection parameters by learning from historical network data. Instead of fixed manual thresholds, the machine learning models continuously optimize detection sensitivity and specificity by adapting to changing network conditions, traffic patterns, and seasonal variations, thereby reducing false alarms while maintaining high detection accuracy.
2Ease of operation
If fixed thresholds are set for anomaly detection, then the detection process is straightforward, but the thresholds become erroneous when traffic density or usage patterns change
Solution Approach 1:
The patent implements dynamic threshold adjustment through machine learning models that continuously learn from incoming network data. The system adapts detection parameters in real-time based on changing traffic density, usage patterns, and network conditions, replacing static fixed thresholds with dynamic adaptive thresholds that evolve with the network environment.
Solution Approach 2:
The system incorporates feedback loops where detection results and ground truth data are fed back into the machine learning models for continuous retraining and optimization. This feedback mechanism allows the system to learn from past performance, adjust to changing conditions, and improve detection accuracy over time, ensuring thresholds remain valid despite network evolution.
3Manufacturing precision
If manual threshold setting is performed for each KPI, then customization is possible, but the process is burdensome and prone to errors
Solution Approach 1:
The patent implements self-service automated threshold generation where machine learning models automatically learn and set optimal thresholds for multiple KPIs without manual intervention. The system autonomously processes historical network data, identifies anomaly patterns, and configures detection parameters across hundreds of thousands of KPIs, eliminating the time-consuming manual configuration process while maintaining or improving detection precision.
Solution Approach 2:
The system employs universal machine learning frameworks that can handle diverse KPI types and network conditions with a single automated process. Instead of requiring separate manual threshold settings for each KPI, the unified ML approach generalizes across different metric types, network segments, and anomaly categories, dramatically reducing configuration time while preserving detection accuracy through adaptive learning.
4Measurement precision
If domain experts perform root cause analysis, then accurate analysis is achieved, but the process is slow and human capital-intensive
Solution Approach 1:
The patent replaces manual expert analysis with automated machine learning-based root cause analysis systems. These systems use supervised learning models trained on historical anomaly data and expert-labeled root causes to automatically classify and identify root causes, substituting human expert labor with intelligent automated classification that maintains high accuracy while dramatically improving analysis speed and scalability.
Solution Approach 2:
The system introduces machine learning models as intermediaries between raw network data and root cause identification. These ML intermediaries process large volumes of network telemetry data, extract relevant features, and map patterns to known root cause categories, enabling rapid automated analysis that preserves the accuracy previously requiring human expert judgment while eliminating the time and resource constraints of manual analysis.
Data Source
AI summary
A method for automated root cause analysis in mobile radio access networks, including: determining mobile radio access network data (e.g., RAN data); detecting an anomaly for a set of user sessions and/or cells from the RAN data; and classifying the detected anomalies using a set of root cause classifiers.