Hybrid Machine Learning Framework for Proactive Fault Detection and Classification in Synchronous Digital Hierarchy (SDH) Networks Using KPI-Based Analysis

F. U. Didigwu *

Department of Electrical and Electronic Engineering, University of Port Harcourt, Rivers State, Nigeria.

J. N. Dike

Department of Electrical and Electronic Engineering, University of Port Harcourt, Rivers State, Nigeria.

*Author to whom correspondence should be addressed.


Abstract

Reliable monitoring of Synchronous Digital Hierarchy (SDH) networks remains important in settings where legacy telecommunications infrastructure is still in operation. This study evaluated a key performance indicator (KPI)-based hybrid machine learning framework for detecting anomalies and classifying faults in SDH network observations collected from a Network Monitoring System in Uyo, Nigeria, between January and June 2026. The indicators examined were Bit Error Rate (BER), Errored Seconds (ES), Severely Errored Seconds (SES), Background Block Errors (BBE), and Unavailable Seconds (UAS). Isolation Forest was used for unsupervised anomaly detection, while Random Forest was used for supervised fault classification. Temporal features, including lagged and rolling statistics, were incorporated to represent changing KPI patterns. The Synthetic Minority Over-sampling Technique (SMOTE) was used to address class imbalance, and Optuna was applied for hyperparameter optimisation. The reported evaluation indicated an overall classification accuracy of 98% and a macro F1-score of 0.90. Reported fault-class precision and recall were 0.81 and 0.80, respectively. The confusion matrix contained 519 true positives, 137 false negatives, 113 false positives, and 9,657 true negatives. The area under the receiver operating characteristic curve was 0.991, while average precision was 0.887. Lagged BBE features and BER were identified as influential predictors. These findings suggest that combining anomaly detection with supervised classification can support KPI-based fault monitoring in the SDH network context examined.

Keywords: SDH, network performance monitoring, fault detection, random forest, isolation forest, SMOTE


How to Cite

Didigwu, F. U., and J. N. Dike. 2026. “Hybrid Machine Learning Framework for Proactive Fault Detection and Classification in Synchronous Digital Hierarchy (SDH) Networks Using KPI-Based Analysis”. Journal of Engineering Research and Reports 28 (11):29-41. https://doi.org/10.9734/jerr/2026/v28i112038.

Downloads

Download data is not yet available.