{"id":{"repo_id":"stellenbosch","oai_identifier":"oai:scholar.sun.ac.za:10019.1/136276"},"canonical_url":"https://search.dev.ndltd.org/etd/stellenbosch/oai:scholar.sun.ac.za:10019.1/136276","repository":{"repo_id":"stellenbosch","name":"Stellenbosch University","base_url":"https://scholar.sun.ac.za/server/oai/request"},"display":{"title":"Predictive Maintenance in Rail Transportation: An Explainable Machine Learning Approach","abstract":"Railway systems are integral to sustainable urban mobility, yet their reliability depends on effective maintenance strategies that prevent costly disruptions and safety hazards. Traditional fixed-interval maintenance approaches, although widely adopted, are inefficient and fail to capitalise on the predictive potential of modern condition-monitoring technologies. Predictive maintenance offers a transformative alternative by enabling data-driven interventions before failures occur. However, implementing predictive maintenance in safety-critical environments requires models that are not only accurate but also interpretable, robust, and validated through rigorous statistical methods. This thesis addresses these challenges by investigating interpretable supervised machine learning techniques for predicting and classifying passenger door failures in Gibela’s X’Trapolis Mega trainset using event-driven data from Alstom’s TrainTracer system. Both failure and non-failure incidences are analysed under time-aware partitions to learn discriminative patterns without temporal leakage. Passenger doors represent a high-risk subsystem due to their frequent actuation, exposure to environmental stressors, and direct impact on operational safety. Failures in this subsystem have the potential to immobilise trains, cause extended dwell times, and compromise passenger security, underscoring the need for proactive monitoring. A structured modelling methodology is proposed, encompassing data understanding, preparation, and model development, supported by advanced evaluation and explainability components. The study evaluates a diverse set of supervised learning algorithms, including decision trees, tree-based ensemble methods, logistic regression, support vector machine approximator’s, and multi-layer perceptron’s. Hyperparameter tuning is performed using time series cross-validation to preserve chronological integrity, while threshold optimisation enhances classification performance under severe class imbalance. Performance is assessed using complementary metrics such as F1-score, precision-recall area under the curve, and precision-recall trade-off’s, ensuring robust evaluation in imbalanced contexts. Statistical verification using the Mann-Whitney U test validates performance differences across folds and between models, providing confidence in model selection. Results show that the tree based ensemble outperform non-linear and neural network approaches. Explainability techniques, including feature importance, permutation importance, and SHapley Additive exPlanations, are integrated to deliver global and local interpretability. These insights enable the derivation of actionable threshold-based rules for maintenance alerts, bridging the gap between predictive modelling and operational decision-making. The findings demonstrate that machine learning models, combined with rigorous evaluation and explainability, have the potential to enhance predictive maintenance strategies, reduce downtime, and improve safety in railway systems. This research contributes a replicable methodology for deploying explainable artificial intelligence in industrial contexts, with the aim to support Gibela’s strategic goal of revitalising South Africa’s rail sector through reliable and efficient commuter services.","abstract_html":"Railway systems are integral to sustainable urban mobility, yet their reliability depends on effective maintenance strategies that prevent costly disruptions and safety hazards. Traditional fixed-interval maintenance approaches, although widely adopted, are inefficient and fail to capitalise on the predictive potential of modern condition-monitoring technologies. Predictive maintenance offers a transformative alternative by enabling data-driven interventions before failures occur. However, implementing predictive maintenance in safety-critical environments requires models that are not only accurate but also interpretable, robust, and validated through rigorous statistical methods. This thesis addresses these challenges by investigating interpretable supervised machine learning techniques for predicting and classifying passenger door failures in Gibela’s X’Trapolis Mega trainset using event-driven data from Alstom’s TrainTracer system. Both failure and non-failure incidences are analysed under time-aware partitions to learn discriminative patterns without temporal leakage. Passenger doors represent a high-risk subsystem due to their frequent actuation, exposure to environmental stressors, and direct impact on operational safety. Failures in this subsystem have the potential to immobilise trains, cause extended dwell times, and compromise passenger security, underscoring the need for proactive monitoring. A structured modelling methodology is proposed, encompassing data understanding, preparation, and model development, supported by advanced evaluation and explainability components. The study evaluates a diverse set of supervised learning algorithms, including decision trees, tree-based ensemble methods, logistic regression, support vector machine approximator’s, and multi-layer perceptron’s. Hyperparameter tuning is performed using time series cross-validation to preserve chronological integrity, while threshold optimisation enhances classification performance under severe class imbalance. Performance is assessed using complementary metrics such as F1-score, precision-recall area under the curve, and precision-recall trade-off’s, ensuring robust evaluation in imbalanced contexts. Statistical verification using the Mann-Whitney U test validates performance differences across folds and between models, providing confidence in model selection. Results show that the tree based ensemble outperform non-linear and neural network approaches. Explainability techniques, including feature importance, permutation importance, and SHapley Additive exPlanations, are integrated to deliver global and local interpretability. These insights enable the derivation of actionable threshold-based rules for maintenance alerts, bridging the gap between predictive modelling and operational decision-making. The findings demonstrate that machine learning models, combined with rigorous evaluation and explainability, have the potential to enhance predictive maintenance strategies, reduce downtime, and improve safety in railway systems. This research contributes a replicable methodology for deploying explainable artificial intelligence in industrial contexts, with the aim to support Gibela’s strategic goal of revitalising South Africa’s rail sector through reliable and efficient commuter services.","abstract_has_math":false,"creators":["Mabaso, Nelisa Zamantungwa Sphumelele"],"institution":"Stellenbosch : Stellenbosch University","degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":["Grobler, Jacomine","Bekker, Anriette"],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026-03","date_published":"2026-03","updated_at":"2026-07-24T04:40:12Z","subjects":[],"languages":["en"],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://scholar.sun.ac.za/handle/10019.1/136276","outbound_label":"Repository record","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Grobler, Jacomine","Bekker, Anriette"]},{"key":"dc:contributor.other","label":"Dc Contributor Other","values":["Stellenbosch University. Faculty of Engineering. Dept. of Industrial Engineering."]},{"key":"dc:creator","label":"Author","values":["Mabaso, Nelisa Zamantungwa Sphumelele"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2026-04-30T11:51:25Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2026-04-30T11:51:25Z"]},{"key":"dc:date.issued","label":"Date","values":["2026-03"]},{"key":"dc:publisher","label":"Institution","values":["Stellenbosch : Stellenbosch University"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://scholar.sun.ac.za/handle/10019.1/136276"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Thesis (MEng)--Stellenbosch University, 2026.","Mabaso, N. Z. S. 2026. Predictive Maintenance in Rail Transportation: An Explainable Machine Learning Approach. Unpublished masters thesis. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/c6cb6f1b-6bb7-42a6-8669-8d3bd46f6392"]},{"key":"dc:description.abstract","label":"Abstract","values":["Railway systems are integral to sustainable urban mobility, yet their reliability depends on effective maintenance strategies that prevent costly disruptions and safety hazards. Traditional fixed-interval maintenance approaches, although widely adopted, are inefficient and fail to capitalise on the predictive potential of modern condition-monitoring technologies. Predictive maintenance offers a transformative alternative by enabling data-driven interventions before failures occur. However, implementing predictive maintenance in safety-critical environments requires models that are not only accurate but also interpretable, robust, and validated through rigorous statistical methods. This thesis addresses these challenges by investigating interpretable supervised machine learning techniques for predicting and classifying passenger door failures in Gibela’s X’Trapolis Mega trainset using event-driven data from Alstom’s TrainTracer system. Both failure and non-failure incidences are analysed under time-aware partitions to learn discriminative patterns without temporal leakage. Passenger doors represent a high-risk subsystem due to their frequent actuation, exposure to environmental stressors, and direct impact on operational safety. Failures in this subsystem have the potential to immobilise trains, cause extended dwell times, and compromise passenger security, underscoring the need for proactive monitoring. A structured modelling methodology is proposed, encompassing data understanding, preparation, and model development, supported by advanced evaluation and explainability components. The study evaluates a diverse set of supervised learning algorithms, including decision trees, tree-based ensemble methods, logistic regression, support vector machine approximator’s, and multi-layer perceptron’s. Hyperparameter tuning is performed using time series cross-validation to preserve chronological integrity, while threshold optimisation enhances classification performance under severe class imbalance. Performance is assessed using complementary metrics such as F1-score, precision-recall area under the curve, and precision-recall trade-off’s, ensuring robust evaluation in imbalanced contexts. Statistical verification using the Mann-Whitney U test validates performance differences across folds and between models, providing confidence in model selection. Results show that the tree based ensemble outperform non-linear and neural network approaches. Explainability techniques, including feature importance, permutation importance, and SHapley Additive exPlanations, are integrated to deliver global and local interpretability. These insights enable the derivation of actionable threshold-based rules for maintenance alerts, bridging the gap between predictive modelling and operational decision-making. The findings demonstrate that machine learning models, combined with rigorous evaluation and explainability, have the potential to enhance predictive maintenance strategies, reduce downtime, and improve safety in railway systems. This research contributes a replicable methodology for deploying explainable artificial intelligence in industrial contexts, with the aim to support Gibela’s strategic goal of revitalising South Africa’s rail sector through reliable and efficient commuter services."]},{"key":"dc:title","label":"Title","values":["Predictive Maintenance in Rail Transportation: An Explainable Machine Learning Approach"]}]}],"canonical_facts":{"dc:contributor.advisor":["Grobler, Jacomine","Bekker, Anriette"],"dc:contributor.other":["Stellenbosch University. Faculty of Engineering. Dept. of Industrial Engineering."],"dc:creator":["Mabaso, Nelisa Zamantungwa Sphumelele"],"dc:date.accessioned":["2026-04-30T11:51:25Z"],"dc:date.available":["2026-04-30T11:51:25Z"],"dc:date.issued":["2026-03"],"dc:description":["Thesis (MEng)--Stellenbosch University, 2026.","Mabaso, N. Z. S. 2026. Predictive Maintenance in Rail Transportation: An Explainable Machine Learning Approach. Unpublished masters thesis. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/c6cb6f1b-6bb7-42a6-8669-8d3bd46f6392"],"dc:description.abstract":["Railway systems are integral to sustainable urban mobility, yet their reliability depends on effective maintenance strategies that prevent costly disruptions and safety hazards. Traditional fixed-interval maintenance approaches, although widely adopted, are inefficient and fail to capitalise on the predictive potential of modern condition-monitoring technologies. Predictive maintenance offers a transformative alternative by enabling data-driven interventions before failures occur. However, implementing predictive maintenance in safety-critical environments requires models that are not only accurate but also interpretable, robust, and validated through rigorous statistical methods. This thesis addresses these challenges by investigating interpretable supervised machine learning techniques for predicting and classifying passenger door failures in Gibela’s X’Trapolis Mega trainset using event-driven data from Alstom’s TrainTracer system. Both failure and non-failure incidences are analysed under time-aware partitions to learn discriminative patterns without temporal leakage. Passenger doors represent a high-risk subsystem due to their frequent actuation, exposure to environmental stressors, and direct impact on operational safety. Failures in this subsystem have the potential to immobilise trains, cause extended dwell times, and compromise passenger security, underscoring the need for proactive monitoring. A structured modelling methodology is proposed, encompassing data understanding, preparation, and model development, supported by advanced evaluation and explainability components. The study evaluates a diverse set of supervised learning algorithms, including decision trees, tree-based ensemble methods, logistic regression, support vector machine approximator’s, and multi-layer perceptron’s. Hyperparameter tuning is performed using time series cross-validation to preserve chronological integrity, while threshold optimisation enhances classification performance under severe class imbalance. Performance is assessed using complementary metrics such as F1-score, precision-recall area under the curve, and precision-recall trade-off’s, ensuring robust evaluation in imbalanced contexts. Statistical verification using the Mann-Whitney U test validates performance differences across folds and between models, providing confidence in model selection. Results show that the tree based ensemble outperform non-linear and neural network approaches. Explainability techniques, including feature importance, permutation importance, and SHapley Additive exPlanations, are integrated to deliver global and local interpretability. These insights enable the derivation of actionable threshold-based rules for maintenance alerts, bridging the gap between predictive modelling and operational decision-making. The findings demonstrate that machine learning models, combined with rigorous evaluation and explainability, have the potential to enhance predictive maintenance strategies, reduce downtime, and improve safety in railway systems. This research contributes a replicable methodology for deploying explainable artificial intelligence in industrial contexts, with the aim to support Gibela’s strategic goal of revitalising South Africa’s rail sector through reliable and efficient commuter services."],"dc:identifier.uri":["https://scholar.sun.ac.za/handle/10019.1/136276"],"dc:language.iso":["en"],"dc:publisher":["Stellenbosch : Stellenbosch University"],"dc:title":["Predictive Maintenance in Rail Transportation: An Explainable Machine Learning Approach"],"dc:type":["Thesis"]},"updated_at":"2026-07-24T04:40:12Z"}