{"id":{"repo_id":"passau-thes","oai_identifier":"oai:kobv.de-opus4-uni-passau:1556"},"canonical_url":"https://search.dev.ndltd.org/etd/passau-thes/oai:kobv.de-opus4-uni-passau:1556","repository":{"repo_id":"passau-thes","name":"Universität Passau","base_url":"https://opus4.kobv.de/opus4-uni-passau/oai"},"display":{"title":"Using Context for Credit Card Fraud Detection","abstract":"Online payment fraud is one of the biggest challenges accompanying the ubiquitous adoption of digital payment methods. The academic literature shows that providing data-driven models with additional context of a transaction results in significant improvements in fraud detection performance. However, the methods used to generate suitable context representations often rely on human expert knowledge, which is expensive and suffers from several limitations. In this thesis, we propose different methods to automate this process by learning these context representations end-to-end on the fraud detection objective. Each of these methods is evaluated on millions of real-world transactions from Worldline, our industrial partner. Central to this thesis is our proposal of the Neural Aggregate Generator (NAG), a neural network that learns context representations automatically. The architecture of the NAG is designed to resemble the structure of expert feature aggregates, while also addressing their limitations. Our evaluation of the NAG reveals that it outperforms both approaches that use expert aggregates and other end-to-end methods across several months of testing. A thorough evaluation shows that the NAG improves over other approaches on several key factors including model size and robustness to shorter sequences. We propose several extensions to the NAG with the dual motive of improved alignment with expert aggregates and improved expressiveness. Our evaluation of these extensions shows comparable performance to the NAG with ancillary benefits in terms of prospective interpretability and model size. We also introduce the novel paradigm of using \\lq future' transactions as context. Our analysis of real-world data from Worldline shows that verification of transactions are often delayed by several days and that within this delay there are often several transactions booked on the card which can be used as additional context. We show that this future context improves the performance of sequence models. Moreover, we also show that a balance between past and future context yields the best results and that using future context allows the use of shorter sequences overall. Beyond context-based fraud detection, we also provide an initial proposal of generating synthetic credit card data using Generative Adversarial Networks (GANs), showing that a Wasserstein GAN can be used to generated synthetic data similar to a popular publicly available credit card fraud dataset. We also describe several possible directions for future work including the incorporation of a adapted self-attention mechanism to the NAG and the use of transformers for synthetic data generation.","abstract_html":"Online payment fraud is one of the biggest challenges accompanying the ubiquitous adoption of digital payment methods. The academic literature shows that providing data-driven models with additional context of a transaction results in significant improvements in fraud detection performance. However, the methods used to generate suitable context representations often rely on human expert knowledge, which is expensive and suffers from several limitations. In this thesis, we propose different methods to automate this process by learning these context representations end-to-end on the fraud detection objective. Each of these methods is evaluated on millions of real-world transactions from Worldline, our industrial partner. Central to this thesis is our proposal of the Neural Aggregate Generator (NAG), a neural network that learns context representations automatically. The architecture of the NAG is designed to resemble the structure of expert feature aggregates, while also addressing their limitations. Our evaluation of the NAG reveals that it outperforms both approaches that use expert aggregates and other end-to-end methods across several months of testing. A thorough evaluation shows that the NAG improves over other approaches on several key factors including model size and robustness to shorter sequences. We propose several extensions to the NAG with the dual motive of improved alignment with expert aggregates and improved expressiveness. Our evaluation of these extensions shows comparable performance to the NAG with ancillary benefits in terms of prospective interpretability and model size. We also introduce the novel paradigm of using \\lq future' transactions as context. Our analysis of real-world data from Worldline shows that verification of transactions are often delayed by several days and that within this delay there are often several transactions booked on the card which can be used as additional context. We show that this future context improves the performance of sequence models. Moreover, we also show that a balance between past and future context yields the best results and that using future context allows the use of shorter sequences overall. Beyond context-based fraud detection, we also provide an initial proposal of generating synthetic credit card data using Generative Adversarial Networks (GANs), showing that a Wasserstein GAN can be used to generated synthetic data similar to a popular publicly available credit card fraud dataset. We also describe several possible directions for future work including the incorporation of a adapted self-attention mechanism to the NAG and the use of transformers for synthetic data generation.","abstract_has_math":false,"creators":["Ghosh Dastidar, Kanishka"],"institution":"Universität Passau","degree_name":null,"degree_level":"thesis.doctoral","degree_discipline":null,"degree_department":null,"school":null,"contributors":["Granitzer, Michael","Habrard, Amaury"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025-02-04","date_published":"2025-02-04","updated_at":"2026-07-24T03:45:10Z","subjects":["Credit Card Fraud Detection","Deep Learning","Neural Networks"],"languages":[],"rights":["Creative Commons - CC BY - Namensnennung 4.0 International"],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://opus4.kobv.de/opus4-uni-passau/frontdoor/index/index/docId/1556","outbound_label":"Repository record","outbound_source":"source_url"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Granitzer, Michael","Habrard, Amaury"]},{"key":"dc:creator","label":"Author","values":["Ghosh Dastidar, Kanishka"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:publisher","label":"Institution","values":["Universität Passau"]},{"key":"dc:type","label":"Dc Type","values":["doctoralThesis"]},{"key":"thesis:degree_level","label":"Degree Level","values":["thesis.doctoral"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["Universität Passau"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Credit Card Fraud Detection","Deep Learning","Neural Networks"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:rights","label":"Dc Rights","values":["Creative Commons - CC BY - Namensnennung 4.0 International"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Online payment fraud is one of the biggest challenges accompanying the ubiquitous adoption of digital payment methods. The academic literature shows that providing data-driven models with additional context of a transaction results in significant improvements in fraud detection performance. However, the methods used to generate suitable context representations often rely on human expert knowledge, which is expensive and suffers from several limitations. In this thesis, we propose different methods to automate this process by learning these context representations end-to-end on the fraud detection objective. Each of these methods is evaluated on millions of real-world transactions from Worldline, our industrial partner. Central to this thesis is our proposal of the Neural Aggregate Generator (NAG), a neural network that learns context representations automatically. The architecture of the NAG is designed to resemble the structure of expert feature aggregates, while also addressing their limitations. Our evaluation of the NAG reveals that it outperforms both approaches that use expert aggregates and other end-to-end methods across several months of testing. A thorough evaluation shows that the NAG improves over other approaches on several key factors including model size and robustness to shorter sequences. We propose several extensions to the NAG with the dual motive of improved alignment with expert aggregates and improved expressiveness. Our evaluation of these extensions shows comparable performance to the NAG with ancillary benefits in terms of prospective interpretability and model size. We also introduce the novel paradigm of using \\lq future' transactions as context. Our analysis of real-world data from Worldline shows that verification of transactions are often delayed by several days and that within this delay there are often several transactions booked on the card which can be used as additional context. We show that this future context improves the performance of sequence models. Moreover, we also show that a balance between past and future context yields the best results and that using future context allows the use of shorter sequences overall. Beyond context-based fraud detection, we also provide an initial proposal of generating synthetic credit card data using Generative Adversarial Networks (GANs), showing that a Wasserstein GAN can be used to generated synthetic data similar to a popular publicly available credit card fraud dataset. We also describe several possible directions for future work including the incorporation of a adapted self-attention mechanism to the NAG and the use of transformers for synthetic data generation.","Betrug bei Online-Zahlungen ist eine der größten Herausforderungen bei der weitverbreiteten Einführung digitaler Zahlungsmethoden. Die akademische Literatur zeigt, dass die Bereitstellung von datengesteuerten Modellen mit zusätzlichem Transaktionskontext zu einer erheblichen Verbesserung der Betrugserkennungsleistung führt. Die Methoden zur Generierung geeigneter Kontextrepräsentationen beruhen jedoch häufig auf menschlichem Expertenwissen, was teuer ist und einige Einschränkungen mit sich bringt. In dieser Arbeit schlagen wir verschiedene Methoden vor, um diesen Prozess zu automatisieren, indem diese Kontextrepräsentationen end-to-end auf das Ziel der Betrugserkennung hin erlernt werden. Jede dieser Methoden wird anhand von Millionen echter Transaktionen von Worldline, unserem industriellen Partner, bewertet. Zentral in dieser Arbeit ist unser Vorschlag des Neural Aggregate Generator (NAG), ein neuronales Netzwerk, das Kontextrepräsentationen automatisch erlernt. Die Architektur des NAGs ist so konzipiert, dass sie der Struktur von Expertenaggregaten ähnelt und gleichzeitig deren Einschränkungen adressiert. Unsere Bewertung des NAGs zeigt, dass es sowohl Ansätze, die Expertenaggregate verwenden, als auch andere end-to-end Methoden in mehreren Monaten des Testens übertrifft. Eine tiefergehende Bewertung zeigt, dass das NAG in mehreren wichtigen Faktoren, einschließlich der Modellgröße und der Robustheit gegenüber kürzeren Sequenzen, bessere Ergebnisse erzielt als andere Ansätze. Wir schlagen mehrere Erweiterungen des NAGs vor, die das doppelte Ziel verfolgen, eine bessere Übereinstimmung mit Expertenaggregaten und eine verbesserte Ausdruckskraft zu erreichen. Unsere Bewertung dieser Erweiterungen zeigt eine vergleichbare Leistung zum NAG mit zusätzlichen Vorteilen hinsichtlich der zukünftigen Interpretierbarkeit und der Modellgröße. Wir führen auch das neuartige Paradigma der Nutzung zukünftiger Transaktionen als Kontext ein. Unsere Analyse der Daten von Worldline zeigt, dass die Verifizierung von Transaktionen oft um mehrere Tage verzögert ist und dass während dieser Verzögerung oft mehrere Transaktionen auf der Karte gebucht werden, die als zusätzlicher Kontext verwendet werden können. Wir zeigen, dass dieser „zukünftige“ Kontext die Leistung von Sequenzmodellen verbessert. Darüber hinaus zeigen wir, dass ein Gleichgewicht zwischen vergangenem und zukünftigem Kontext die besten Ergebnisse liefert und dass die Nutzung zukünftigen Kontexts insgesamt kürzere Sequenzen ermöglicht. Über die kontextbasierte Betrugserkennung hinaus geben wir auch einen ersten Vorschlag zur Generierung synthetischer Kreditkartendaten mittels Generative Adversarial Networks (GANs) und zeigen, dass ein durch einen Wasserstein-GAN generierter Datensatz geeignet ist, um darauf Erkennungsmodelle zu trainieren. Wir beschreiben auch mehrere mögliche Richtungen für zukünftige Arbeiten, einschließlich der Integration eines angepassten Self-Attention Mechanismus in das NAG und der Verwendung von Transformern zur Generierung synthetischer Daten."]},{"key":"dc:format.medium","label":"Dc Format Medium","values":["application/pdf"]},{"key":"dc:title","label":"Title","values":["Using Context for Credit Card Fraud Detection"]}]}],"canonical_facts":{"dc:contributor":["Granitzer, Michael","Habrard, Amaury"],"dc:creator":["Ghosh Dastidar, Kanishka"],"dc:description.abstract":["Online payment fraud is one of the biggest challenges accompanying the ubiquitous adoption of digital payment methods. The academic literature shows that providing data-driven models with additional context of a transaction results in significant improvements in fraud detection performance. However, the methods used to generate suitable context representations often rely on human expert knowledge, which is expensive and suffers from several limitations. In this thesis, we propose different methods to automate this process by learning these context representations end-to-end on the fraud detection objective. Each of these methods is evaluated on millions of real-world transactions from Worldline, our industrial partner. Central to this thesis is our proposal of the Neural Aggregate Generator (NAG), a neural network that learns context representations automatically. The architecture of the NAG is designed to resemble the structure of expert feature aggregates, while also addressing their limitations. Our evaluation of the NAG reveals that it outperforms both approaches that use expert aggregates and other end-to-end methods across several months of testing. A thorough evaluation shows that the NAG improves over other approaches on several key factors including model size and robustness to shorter sequences. We propose several extensions to the NAG with the dual motive of improved alignment with expert aggregates and improved expressiveness. Our evaluation of these extensions shows comparable performance to the NAG with ancillary benefits in terms of prospective interpretability and model size. We also introduce the novel paradigm of using \\lq future' transactions as context. Our analysis of real-world data from Worldline shows that verification of transactions are often delayed by several days and that within this delay there are often several transactions booked on the card which can be used as additional context. We show that this future context improves the performance of sequence models. Moreover, we also show that a balance between past and future context yields the best results and that using future context allows the use of shorter sequences overall. Beyond context-based fraud detection, we also provide an initial proposal of generating synthetic credit card data using Generative Adversarial Networks (GANs), showing that a Wasserstein GAN can be used to generated synthetic data similar to a popular publicly available credit card fraud dataset. We also describe several possible directions for future work including the incorporation of a adapted self-attention mechanism to the NAG and the use of transformers for synthetic data generation.","Betrug bei Online-Zahlungen ist eine der größten Herausforderungen bei der weitverbreiteten Einführung digitaler Zahlungsmethoden. Die akademische Literatur zeigt, dass die Bereitstellung von datengesteuerten Modellen mit zusätzlichem Transaktionskontext zu einer erheblichen Verbesserung der Betrugserkennungsleistung führt. Die Methoden zur Generierung geeigneter Kontextrepräsentationen beruhen jedoch häufig auf menschlichem Expertenwissen, was teuer ist und einige Einschränkungen mit sich bringt. In dieser Arbeit schlagen wir verschiedene Methoden vor, um diesen Prozess zu automatisieren, indem diese Kontextrepräsentationen end-to-end auf das Ziel der Betrugserkennung hin erlernt werden. Jede dieser Methoden wird anhand von Millionen echter Transaktionen von Worldline, unserem industriellen Partner, bewertet. Zentral in dieser Arbeit ist unser Vorschlag des Neural Aggregate Generator (NAG), ein neuronales Netzwerk, das Kontextrepräsentationen automatisch erlernt. Die Architektur des NAGs ist so konzipiert, dass sie der Struktur von Expertenaggregaten ähnelt und gleichzeitig deren Einschränkungen adressiert. Unsere Bewertung des NAGs zeigt, dass es sowohl Ansätze, die Expertenaggregate verwenden, als auch andere end-to-end Methoden in mehreren Monaten des Testens übertrifft. Eine tiefergehende Bewertung zeigt, dass das NAG in mehreren wichtigen Faktoren, einschließlich der Modellgröße und der Robustheit gegenüber kürzeren Sequenzen, bessere Ergebnisse erzielt als andere Ansätze. Wir schlagen mehrere Erweiterungen des NAGs vor, die das doppelte Ziel verfolgen, eine bessere Übereinstimmung mit Expertenaggregaten und eine verbesserte Ausdruckskraft zu erreichen. Unsere Bewertung dieser Erweiterungen zeigt eine vergleichbare Leistung zum NAG mit zusätzlichen Vorteilen hinsichtlich der zukünftigen Interpretierbarkeit und der Modellgröße. Wir führen auch das neuartige Paradigma der Nutzung zukünftiger Transaktionen als Kontext ein. Unsere Analyse der Daten von Worldline zeigt, dass die Verifizierung von Transaktionen oft um mehrere Tage verzögert ist und dass während dieser Verzögerung oft mehrere Transaktionen auf der Karte gebucht werden, die als zusätzlicher Kontext verwendet werden können. Wir zeigen, dass dieser „zukünftige“ Kontext die Leistung von Sequenzmodellen verbessert. Darüber hinaus zeigen wir, dass ein Gleichgewicht zwischen vergangenem und zukünftigem Kontext die besten Ergebnisse liefert und dass die Nutzung zukünftigen Kontexts insgesamt kürzere Sequenzen ermöglicht. Über die kontextbasierte Betrugserkennung hinaus geben wir auch einen ersten Vorschlag zur Generierung synthetischer Kreditkartendaten mittels Generative Adversarial Networks (GANs) und zeigen, dass ein durch einen Wasserstein-GAN generierter Datensatz geeignet ist, um darauf Erkennungsmodelle zu trainieren. Wir beschreiben auch mehrere mögliche Richtungen für zukünftige Arbeiten, einschließlich der Integration eines angepassten Self-Attention Mechanismus in das NAG und der Verwendung von Transformern zur Generierung synthetischer Daten."],"dc:format.medium":["application/pdf"],"dc:publisher":["Universität Passau"],"dc:rights":["Creative Commons - CC BY - Namensnennung 4.0 International"],"dc:subject":["Credit Card Fraud Detection","Deep Learning","Neural Networks"],"dc:title":["Using Context for Credit Card Fraud Detection"],"dc:type":["doctoralThesis"],"thesis:degree_level":["thesis.doctoral"],"thesis:institution_name":["Universität Passau"]},"updated_at":"2026-07-24T03:45:10Z"}