RAN ML Model Input Data Imputation Using Modal Value Replacement

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Next Generation mobile networks, such as 5G, face challenges in providing high-bandwidth, low-latency services due to data quality issues like missing values and skewed distributions in machine learning (ML) models used for predictive analytics, which affect the accuracy and consistency of network performance predictions.

Innovation Solution

The implementation of modal value replacement for null values in ML models, particularly suited for skewed distributions, improves data curation and reliability, enhancing the consistency and accuracy of ML input datasets for RAN performance predictions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional data handling methods are used for ML models in RAN networks, then the implementation is simple, but the accuracy and consistency of network performance predictions deteriorate due to missing values and skewed distributions

Engineering Contradiction:
Improveprediction accuracyVSAvoiddata processing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies preliminary action by performing data imputation and transformation before feeding data into ML models. Specifically, missing values are imputed using statistical methods (mean, median, or mode imputation) and skewed distributions are transformed using log or square root transformations prior to model training, thereby improving prediction accuracy without adding complex processing during model execution

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent changes data parameters by applying statistical transformations to handle skewed distributions. The system detects skewed features and automatically applies appropriate transformations (log transformation, square root transformation, or box-cox transformation) to normalize the data distribution, thereby improving ML model performance on network performance predictions

Inventive Principle:
Principle #35Parameter changes

2Reliability

If modal value replacement is used for null values in skewed distributions, then data reliability improves, but the processing time increases

Engineering Contradiction:
Improvedata reliabilityVSAvoiddata processing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent applies partial action by selectively imputing only the most critical or numerous missing values using modal value replacement, rather than attempting to impute every single missing value. The system identifies features with high missing value proportions and applies imputation primarily to those features, thereby improving data reliability while minimizing additional processing time

Inventive Principle:
Principle #16Partial or excessive action

Solution Approach 2:

The system performs self-service by automatically detecting missing values and applying appropriate imputation methods without requiring manual intervention. The data preprocessing pipeline automatically identifies null values, selects the most appropriate imputation strategy based on data characteristics, and executes the imputation process autonomously, reducing both time loss and operational complexity

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP3895376B1System and method for improving machine learning model performance in a communications network
Publication Date: 2023.10.18 TELEFONAKTIEBOLAGET LM ERICSSON (PUBL)
  • EP3895376B1 patent drawingFigure 1
  • EP3895376B1 patent drawingFigure 2A
  • EP3895376B1 patent drawingFigure 2B

AI summary

A system, method and non-transitory computer readable media for optimizing input data for an ML model associated with a communications network (100). In one implementation, example ML model(s) may be trained using a modified dataset 5 obtained for a plurality of cellular aggregation units (CAUs) (304-1 to 304-N) of the RAN infrastructure(s) (102, 302), wherein the modified dataset is derived from data collected for individual CAUs (304-1 to 304-N) over a data collection period with respect to a plurality of KPI variables. The modified data set is optimized by replacement of null values of variables with corresponding modal values of the 10 variables. The trained ML model may be used for predicting one or more KPIs based on a set of test data associated with the RAN infrastructure(s) (102, 302).