TY - JOUR A1 - Weißgerber, Thomas A1 - Ben Amor, Mehdi A1 - Fellicious, Christofer A1 - Granitzer, Michael T1 - PyPads: Transparent Machine Learning Experiment Tracking JF - Datenbank-Spektrum N2 - Despite algorithmic advancements in the field of machine learning, a need for improvement in the infrastructure supporting machine learning development and research has become increasingly apparent. Machine learning experiments usually tend to be more ad-hoc in nature, and results are communicated most often in the form of a publication. Experimental details are often omitted due to size or time constraints, or simply because the complexity in terms of technical setup or parametrization became intractable. Even access to code bases, disregard important properties of the environment and experimental setup, like for example random generators or computing infrastructure. At the same time, tracking and communicating an often inherently exploratory scientific process is a task with considerable effort. We explored different venues to tackle these issues from a data science engineering point of view. The efforts resulted in PyPads, a framework providing an infrastructure to extend experimental setups with logging, communication and analysis features in a mostly non-intrusive way. PyPads can be extended to different Python-based frameworks, utilizing community driven, descriptive metadata in an effort to harmonize library specific logs in an ontology. Meanwhile, we also try to emphasize similarities to practices in software engineering, which have turned out to be essential in practical applications. KW - Machine Learning KW - Reproducibility KW - Open Science KW - Automated Logging KW - Python Y1 - 2023 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:101:1-2024021511222778687954 VL - 24 IS - 1 SP - 53 EP - 62 PB - Springer Nature CY - Berlin ER - TY - THES A1 - Stoffels, Dominik T1 - Advancing Pattern Detection, Theory Development and Decision Making with Explainable AI N2 - The application of explainable artificial intelligence (XAI) methods in data-driven decision-making and computationally intensive theory development (CTD) is a subject of ongoing debate, particularly concerning how and whether these methods can be effectively employed, and how the reliability of their explanations can be ensured. This dissertation addresses these issues by systematically analyzing the usability of XAI for pattern detection, CTD, and decision-making, drawing on various real-world and synthetic datasets and employing different empirical methods and perspectives. The dissertation consists of four studies, each addressing distinct issues in the field of XAI application. KW - Explainable Artificial Intelligence KW - Machine Learning KW - Computationally Intensive Theory Development Y1 - 2025 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:739-opus4-15975 ER - TY - THES A1 - Jurgovsky, Johannes T1 - Context-Aware Credit Card Fraud Detection N2 - Credit card fraud has emerged as major problem in the electronic payment sector. In this thesis, we study data-driven fraud detection and address several of its intricate challenges by means of machine learning methods with the goal to identify fraudulent transactions that have been issued illegitimately on behalf of the rightful card owner. In particular, we explore several means to leverage contextual information beyond a transaction’s basic attributes on the transaction level, sequence level and user level. On the transaction level, we aim to identify fraudulent transactions which, in terms of their attribute values, are globally distinguishable from genuine transactions. We provide an empirical study of the influence of class imbalance and forecasting horizons on the classification performance of a random forest classifier. We augment transactions with additional features extracted from external knowledge sources and show that external information about countries and calendar events improves classification performance most noticeably on card-not-present transactions. On the sequence level, we aim to detect frauds that are inconspicuous in the background of all transactions but peculiar with respect to the short-term sequence they appear in. We use a Long Short-term Memory network (LSTM) for modeling the sequential succession of transactions. Our results suggest that LSTM-based modeling is a promising strategy for characterizing sequences of card-present transactions but it is not adequate for card-not-present transactions. On the user level, we elaborate on feature aggregations and propose a flexible concept allowing us define numerous features by means of a simple syntax. We provide a CUDA-based implementation for the computationally expensive extraction with a speed-up of two orders of magnitude over a single-core implementation. Our feature selection study reveals that aggregates extracted from users’ transaction sequences are more useful than those extracted from merchant sequences. Moreover, we discover multiple sets of candidate features with equivalent performance as manually engineered aggregates while being structurally different. Regarding future work, we motivate the usage of simple and transparent machine learning methods for credit card fraud detection and we sketch a simple user-focused modeling approach. KW - Credit Card Fraud Detection KW - Machine Learning KW - Data Augmentation KW - Feature Engineering KW - Kreditkartenmissbrauch KW - Computersicherheit Y1 - 2019 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:739-opus4-7622 ER -