{"id":{"repo_id":"oxford-brookes","oai_identifier":"tle:90db0878-5322-42e9-b3e4-d8dcc9d9da84:d6bd9758-527a-46cd-bfe2-c433766e8fca:1"},"canonical_url":"https://search.dev.ndltd.org/etd/oxford-brookes/tle:90db0878-5322-42e9-b3e4-d8dcc9d9da84:d6bd9758-527a-46cd-bfe2-c433766e8fca:1","repository":{"repo_id":"oxford-brookes","name":"Oxford Brookes University","base_url":"https://radar.brookes.ac.uk/radar/oai"},"display":{"title":"k-Label Dependent Evolutionary Distance Weighting for Software Defect Prediction","abstract":"Software Defect Prediction is a field of study that uses machine learning algorithms to identify software that is susceptible to defects. Software defects and bugs can potentially cause errors that have significant consequences ranging from minor inconveniences to system failures. Identification of errors at an early stage allows software developers to reduce resources used and to provide more reliable software. Conventional SDP models often depend on static feature-weighting approaches that cannot be adjusted to the particularities of different data sets. Existing model address some of these challenges, but lack interpretability or prediction accuracy. This research addresses the limitations of existing models by introducing an enhanced $k$-Label-Dependent Evolutionary Distance Weighting ($k$ -LDEDW) method, which combines an adaptive genetic algorithm and $k$-NN classifier ($k$-NN) to improve predictions by dynamically adjusting the feature weights for each data point. The proposed adaptive $k$-LDEDW model is implemented to predict software defects and the performance of the model is evaluated in the ApacheJIT data set, which contains unbalanced commits of clean and bug-inducing software. The adaptive genetic algorithm refines the weights using crossover and mutation techniques, guided by Matthew correlation coefficient (MCC) as the fitness function. In addition, Shapley values are used to explain the importance of software features in defect predictions to provide insights into the interpretability of the model. The $k$-LDEDW model is then compared with graph neural networks (GNN), clustering techniques, and JIT bug prediction models. The respective models are compared using evaluation metrics for accuracy, precision, recall, $f_1$-score, AUC, and MCC. The results show that $k$-LDEDW outperformed other methods, achieving an MCC of 0.65, precision of 0.86, recall of 0.98, and an $f_1$-score of 0.92 for clean commits. For buggy commits, it attained a precision of 0.91, while the overall predictions yielded a precision of 0.88 and recall of 0.76. Shapley values revealed that $k$-LDEDW utilized a diverse range of features, showing that the model offers a holistic evaluation of software features. These results highlight the robustness of $k$-LDEDW and its ability to adapt to an imbalanced data set.","abstract_html":"Software Defect Prediction is a field of study that uses machine learning algorithms to identify software that is susceptible to defects. Software defects and bugs can potentially cause errors that have significant consequences ranging from minor inconveniences to system failures. Identification of errors at an early stage allows software developers to reduce resources used and to provide more reliable software. Conventional SDP models often depend on static feature-weighting approaches that cannot be adjusted to the particularities of different data sets. Existing model address some of these challenges, but lack interpretability or prediction accuracy. This research addresses the limitations of existing models by introducing an enhanced $k$-Label-Dependent Evolutionary Distance Weighting ($k$ -LDEDW) method, which combines an adaptive genetic algorithm and $k$-NN classifier ($k$-NN) to improve predictions by dynamically adjusting the feature weights for each data point. The proposed adaptive $k$-LDEDW model is implemented to predict software defects and the performance of the model is evaluated in the ApacheJIT data set, which contains unbalanced commits of clean and bug-inducing software. The adaptive genetic algorithm refines the weights using crossover and mutation techniques, guided by Matthew correlation coefficient (MCC) as the fitness function. In addition, Shapley values are used to explain the importance of software features in defect predictions to provide insights into the interpretability of the model. The $k$-LDEDW model is then compared with graph neural networks (GNN), clustering techniques, and JIT bug prediction models. The respective models are compared using evaluation metrics for accuracy, precision, recall, <span class=\"etd-inline-math\">f<sub>1</sub></span>-score, AUC, and MCC. The results show that $k$-LDEDW outperformed other methods, achieving an MCC of 0.65, precision of 0.86, recall of 0.98, and an <span class=\"etd-inline-math\">f<sub>1</sub></span>-score of 0.92 for clean commits. For buggy commits, it attained a precision of 0.91, while the overall predictions yielded a precision of 0.88 and recall of 0.76. Shapley values revealed that $k$-LDEDW utilized a diverse range of features, showing that the model offers a holistic evaluation of software features. These results highlight the robustness of $k$-LDEDW and its ability to adapt to an imbalanced data set.","abstract_has_math":true,"creators":["Herman, Muhammad Imran Bin"],"institution":"Oxford Brookes University","degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":["Harrison, Rachel","Niri, Manny"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":null,"date_issued":"","date_published":null,"updated_at":"2026-07-24T03:42:15Z","subjects":[],"languages":["en"],"rights":["All rights reserved"],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://doi.org/10.24384/ha27-n664","outbound_label":"DOI","outbound_source":"dc:identifier"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Herman, Muhammad Imran Bin","Harrison, Rachel","Niri, Manny"]},{"key":"dc:creator","label":"Author","values":["Herman, Muhammad Imran Bin"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:publisher","label":"Institution","values":["Oxford Brookes University"]},{"key":"dc:type","label":"Dc Type","values":["thesis"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["en"]},{"key":"dc:rights","label":"Dc Rights","values":["All rights reserved"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://doi.org/10.24384/ha27-n664","https://radar.brookes.ac.uk/radar/file/90db0878-5322-42e9-b3e4-d8dcc9d9da84/1/Herman2025SoftwareDefectPrediction.pdf"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Software Defect Prediction is a field of study that uses machine learning algorithms to identify software that is susceptible to defects. Software defects and bugs can potentially cause errors that have significant consequences ranging from minor inconveniences to system failures. Identification of errors at an early stage allows software developers to reduce resources used and to provide more reliable software. Conventional SDP models often depend on static feature-weighting approaches that cannot be adjusted to the particularities of different data sets. Existing model address some of these challenges, but lack interpretability or prediction accuracy. This research addresses the limitations of existing models by introducing an enhanced $k$-Label-Dependent Evolutionary Distance Weighting ($k$ -LDEDW) method, which combines an adaptive genetic algorithm and $k$-NN classifier ($k$-NN) to improve predictions by dynamically adjusting the feature weights for each data point. The proposed adaptive $k$-LDEDW model is implemented to predict software defects and the performance of the model is evaluated in the ApacheJIT data set, which contains unbalanced commits of clean and bug-inducing software. The adaptive genetic algorithm refines the weights using crossover and mutation techniques, guided by Matthew correlation coefficient (MCC) as the fitness function. In addition, Shapley values are used to explain the importance of software features in defect predictions to provide insights into the interpretability of the model. The $k$-LDEDW model is then compared with graph neural networks (GNN), clustering techniques, and JIT bug prediction models. The respective models are compared using evaluation metrics for accuracy, precision, recall, $f_1$-score, AUC, and MCC. The results show that $k$-LDEDW outperformed other methods, achieving an MCC of 0.65, precision of 0.86, recall of 0.98, and an $f_1$-score of 0.92 for clean commits. For buggy commits, it attained a precision of 0.91, while the overall predictions yielded a precision of 0.88 and recall of 0.76. Shapley values revealed that $k$-LDEDW utilized a diverse range of features, showing that the model offers a holistic evaluation of software features. These results highlight the robustness of $k$-LDEDW and its ability to adapt to an imbalanced data set."]},{"key":"dc:format","label":"Dc Format","values":["application/pdf"]},{"key":"dc:title","label":"Title","values":["k-Label Dependent Evolutionary Distance Weighting for Software Defect Prediction"]}]}],"canonical_facts":{"dc:contributor":["Herman, Muhammad Imran Bin","Harrison, Rachel","Niri, Manny"],"dc:creator":["Herman, Muhammad Imran Bin"],"dc:description":["Software Defect Prediction is a field of study that uses machine learning algorithms to identify software that is susceptible to defects. Software defects and bugs can potentially cause errors that have significant consequences ranging from minor inconveniences to system failures. Identification of errors at an early stage allows software developers to reduce resources used and to provide more reliable software. Conventional SDP models often depend on static feature-weighting approaches that cannot be adjusted to the particularities of different data sets. Existing model address some of these challenges, but lack interpretability or prediction accuracy. This research addresses the limitations of existing models by introducing an enhanced $k$-Label-Dependent Evolutionary Distance Weighting ($k$ -LDEDW) method, which combines an adaptive genetic algorithm and $k$-NN classifier ($k$-NN) to improve predictions by dynamically adjusting the feature weights for each data point. The proposed adaptive $k$-LDEDW model is implemented to predict software defects and the performance of the model is evaluated in the ApacheJIT data set, which contains unbalanced commits of clean and bug-inducing software. The adaptive genetic algorithm refines the weights using crossover and mutation techniques, guided by Matthew correlation coefficient (MCC) as the fitness function. In addition, Shapley values are used to explain the importance of software features in defect predictions to provide insights into the interpretability of the model. The $k$-LDEDW model is then compared with graph neural networks (GNN), clustering techniques, and JIT bug prediction models. The respective models are compared using evaluation metrics for accuracy, precision, recall, $f_1$-score, AUC, and MCC. The results show that $k$-LDEDW outperformed other methods, achieving an MCC of 0.65, precision of 0.86, recall of 0.98, and an $f_1$-score of 0.92 for clean commits. For buggy commits, it attained a precision of 0.91, while the overall predictions yielded a precision of 0.88 and recall of 0.76. Shapley values revealed that $k$-LDEDW utilized a diverse range of features, showing that the model offers a holistic evaluation of software features. These results highlight the robustness of $k$-LDEDW and its ability to adapt to an imbalanced data set."],"dc:format":["application/pdf"],"dc:identifier":["https://doi.org/10.24384/ha27-n664","https://radar.brookes.ac.uk/radar/file/90db0878-5322-42e9-b3e4-d8dcc9d9da84/1/Herman2025SoftwareDefectPrediction.pdf"],"dc:language":["en"],"dc:publisher":["Oxford Brookes University"],"dc:rights":["All rights reserved"],"dc:title":["k-Label Dependent Evolutionary Distance Weighting for Software Defect Prediction"],"dc:type":["thesis"]},"updated_at":"2026-07-24T03:42:15Z"}