Intelligent Computing / 2022 / Article / Tab 1

Research Article

Data-Driven Machine Learning Techniques for Self-Healing in Cellular Wireless Networks: Challenges and Solutions

Table 1

The solutions for different challenges on machine learning in self-healing.

ChallengesSolutionsMethodsBrief description

Data imbalanceData preprocessingOversamplingDuplicating minority class data
UndersamplingRemoving partial majority class data
Oversampling+undersamplingAlleviating the problems caused by using one of them
SMOTE and its variantsCombining resampling and other methods, e.g., KNN
Algorithm modificationOne-classEstimating a boundary in each class
Hybrid strategyCombining learning algorithms with data preprocessing

Data insufficiencyData preprocessingOversamplingDuplicating the samples in each class
SMOTE and its variantsCombining resampling and other methods, e.g., KNN
Using unlabeled dataActive learningSelecting useful unlabeled data and annotating them artificially
Unsupervised learningDealing with the clustering problems of unlabeled data
Semisupervised learningCombining labeled and unlabeled data
Algorithm modificationTransfer learningTransferring learning tasks from target to source domains

Cost insensitivityCost-sensitive learningMisclassification cost-sensitive learningSetting costs for different classes or samples
Introducing new evaluation metrics-measure, -mean, ROC precision-recall, and cost curveCombining different components of confusion matrix to estimate system performance comprehensively

Non-real-time responseProactive responseProactive context self-healingUpgrading existing self-healing from reactive to proactive response

Multisource dataData fusionProbability based methodsObtaining consistent information from random variables, events, or process
The theory of evidence-based methodsUsing symbolic variables and combination rules to infer consistent information from multisource data
Artificial intelligence-based approachesOffering a good solution for large-scale complex data