Comparative Performance Evaluation of Bayesian-Optimized Random Forest and Conventional Machine Learning Techniques for Predictive Maintenance of Nigerian Aging Transmission Lines

Maintaining aging high-voltage transmission assets requires dependable forecasts of failure risk rather than responses based solely on fixed inspection intervals. This paper evaluates Bayesian-optimized Random Forest (BO-RF) alongside conventional Random Forest (RF), gradient boosting, artificial neural network (ANN), support vector machine (SVM) and decision tree classifiers for predicting failures in a subsequent 30-day interval. A reproducible simulation generated 7,200 line-month observations representing 100 hypothetical transmission corridors between 2020 and 2025. Nine condition and exposure variables described asset age, loading, ambient temperature, lightning, corrosion, insulator condition, vegetation clearance, conductor hotspot temperature and preceding faults. The investigation separated the initial training period (2020–2022), Bayesian hyperparameter selection (2023), maintenance-alert threshold determination (2024) and final assessment (2025). Eighteen optimization trials selected RF settings by validation average precision, while alarm thresholds reflected an assumed missed-failure penalty four times the cost of a false alarm. In the 1,200-observation test year, BO-RF attained average precision of 0.316, F1-score of 0.361, recall of 0.465 and ROC-AUC of 0.695. Conventional RF attained average precision of 0.281 and F1-score of 0.335; the ANN achieved the highest ROC-AUC (0.700). At its selected threshold, BO-RF identified 73 failure events, issued 174 false alerts and missed 84 events. Its decision loss was 510 normalized units, compared with 529 for RF. Corridor-level bootstrap analysis supported a positive difference in average precision, although the confidence interval for the reduction in decision loss crossed zero. The investigation shows that evaluation of rare failures depends on temporal validation, false alarm consequences and probability quality, not classification accuracy alone. Because the observations arise from a controlled simulation rather than utility maintenance records, the findings establish a testable methodology, not field-verified performance in Nigeria. Future applications require timestamped Nigerian transmission-line records, calibrated intervention costs and prospective testing. Attention should also address regional climate differences, measurement quality, and maintenance resource constraints.

Authors

Institutions

Publication Details

Journal
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE AND MATHEMATICAL THEORY E-ISSN
Published
2026-11-16
DOI
https://doi.org/10.56201/ijcsmt.vol.12.no10.2026.pg154.166
Primary Topic
Reliability and Maintenance Optimization
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Comparative Performance Evaluation of Bayesian-Optimized Random Forest and Conventional Machine Learning Techniques for Predictive Maintenance of Nigerian Aging Transmission Lines

Ganiyu Adedayo Ajenikoko
2 links
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE AND MATHEMATICAL THEORY E-ISSN
Reliability and Maintenance Optimization
article

Comparative Performance Evaluation of Bayesian-Optimized Random Forest and Conventional Machine Learning Techniques for Predictive Maintenance of Nigerian Aging Transmission Lines

Ganiyu Adedayo Ajenikoko
article en

Abstract

Maintaining aging high-voltage transmission assets requires dependable forecasts of failure risk rather than responses based solely on fixed inspection intervals. This paper evaluates Bayesian-optimized Random Forest (BO-RF) alongside conventional Random Forest (RF), gradient boosting, artificial neural network (ANN), support vector machine (SVM) and decision tree classifiers for predicting failures in a subsequent 30-day interval. A reproducible simulation generated 7,200 line-month observations representing 100 hypothetical transmission corridors between 2020 and 2025. Nine condition and exposure variables described asset age, loading, ambient temperature, lightning, corrosion, insulator condition, vegetation clearance, conductor hotspot temperature and preceding faults. The investigation separated the initial training period (2020–2022), Bayesian hyperparameter selection (2023), maintenance-alert threshold determination (2024) and final assessment (2025). Eighteen optimization trials selected RF settings by validation average precision, while alarm thresholds reflected an assumed missed-failure penalty four times the cost of a false alarm. In the 1,200-observation test year, BO-RF attained average precision of 0.316, F1-score of 0.361, recall of 0.465 and ROC-AUC of 0.695. Conventional RF attained average precision of 0.281 and F1-score of 0.335; the ANN achieved the highest ROC-AUC (0.700). At its selected threshold, BO-RF identified 73 failure events, issued 174 false alerts and missed 84 events. Its decision loss was 510 normalized units, compared with 529 for RF. Corridor-level bootstrap analysis supported a positive difference in average precision, although the confidence interval for the reduction in decision loss crossed zero. The investigation shows that evaluation of rare failures depends on temporal validation, false alarm consequences and probability quality, not classification accuracy alone. Because the observations arise from a controlled simulation rather than utility maintenance records, the findings establish a testable methodology, not field-verified performance in Nigeria. Future applications require timestamped Nigerian transmission-line records, calibrated intervention costs and prospective testing. Attention should also address regional climate differences, measurement quality, and maintenance resource constraints.

INTERNATIONAL JOURNAL OF COMPUTER SCIENCE AND MATHEMATICAL THEORY E-ISSN
Ladoke Akintola University of Technology (NG)
Industry, innovation and infrastructure
Reliability and Maintenance Optimization
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon