Search

Datasets with missing values are very common in industry applications. Missing data typically have a negative impact on machine learning models. With the rise of generative models in deep learning, recent studies proposed solutions to the problem of imputing missing values based various deep generative models. Previous experiments with Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs) showed promising results in this domain. Initially, these results focused on imputation in image data, e.g. filling missing patches in images. Recent proposals addressed missing values in tabular data. For these data, the case for deep generative models seems to be less clear. In the process of providing a fair comparison of proposed methods, we uncover several issues when assessing the status quo: the use of under-specified and ambiguous dataset names, the large range of parameters and hyper-parameters to tune for each method, and the use of different metrics and evaluation methods.

We proposes an algorithm to learn automata infinite alphabets, or at least too large to enumerate. We apply it to define a generic model intended for regression, with transitions constrained by intervals over the alphabet. The algorithm is based on the Red \& Blue framework for learning from an input sample. We show two small case studies where the alphabets are respectively the natural and real numbers, and show how nice properties of automata models like interpretability and graphical representation transfer to regression where typical models are hard to interpret.

We present a Python package for learning (non-)probabilistic deterministic finite state automata and provide heuristics in the red-blue framework. As our package is built along the API of the popular \texttt{scikit-learn} package, it is easy to use and new learning methods are easy to add. It provides PDFA learning as an additional tool for sequence prediction or classification to data scientists, without the need to understand the algorithm itself but rather the limitations of PDFA as a model. With applications of automata learning in diverse fields such as network traffic analysis, software engineering and biology, a stratified package opens opportunities for practitioners.

Ziel der vorliegenden Untersuchung war es, die Konsumgewohnheiten alkoholischer Getränke in einer bislang wenig untersuchten Subpopulation, den Studierenden, zu untersuchen. Zu diesem Zweck wurden Prävalenzen bezüglich der Konsummenge, der Konsummuster, des Rauschtrinkens (Binge-Drinking) sowie alkoholbezogener Störungen, wie Missbrauch oder Abhängigkeit, ermittelt. Des Weiteren wurden spezifische mögliche Einflussfaktoren auf den Alkoholkonsum von Studierenden untersucht. Hierzu wurden demographische Faktoren, die Strukturen des Studiums, insbesondere auch vor dem Hintergrund der Umstellung der Studienabschlüsse gemäß den Bologna-Beschlüsse, sowie allgemeine psychische Belastungen näher betrachtet. Die Erhebung der Daten erfolgte im Jahr 2008 anhand einer Online-Befragung an 2.348 Studierenden dreier niedersächsischer Hochschulen, der TU Braunschweig, der HBK Braunschweig und der Fachhochschule Braunschweig/Wolfenbüttel. Im Anschluss hieran wurde zur Güteabschätzung des Online-Fragebogens und zur spezifischeren Betrachtung pathologischer Trinkmuster und psychischer Auffälligkeiten ein strukturiertes klinisches Interview mit 72 Teilnehmern des Online-Fragebogens durchgeführt. Die Ergebnisse zeigen einen deutlich erhöhten Alkoholkonsum: Nur 10,2% der Studierenden gaben bezogen auf die letzten 30 Tage an, abstinent zu sein, 70,8% tranken risikoarm. 19% konsumierten Alkohol in einem mindestens riskanten Maße. Dieses Ergebnis spiegelt sich auch im Rauschtrinken wider: 34,9% der Studierenden waren Binge-Drinker, weitere 14,5% sog. Heavy User, die sich fünf oder mehr Tage pro Monat rauschmäßig betranken. Nach Auswertung eines Alkoholismus-Screening-Verfahrens (CAGE) betrieben 30,3% mindestens missbräuchlichen, wenn nicht abhängigen Konsum. Die Ergebnisse decken deutlich andere Konsummuster als in der Normalbevölkerung auch gleichen Alters auf. Als Gründe scheinen jedoch weniger internale Bedingungen, wie Coping bei Depressivität oder Ängstlichkeit, sondern vielmehr sozialmotivierte Faktoren eine Rolle zu spielen, die sich zum Teil sogar positiv auf das psychische Wohlbefinden auswirken. Weitere Erklärungsmuster werden abschließend diskutiert., The aim of this paper is to examine the pattern of alcohol use among a sub-group, namely students, which up to this point has been only examined insufficiently. To achieve this, the prevalence of the amount of alcohol consumed, the way in which it is consumed, ‘Binge-drinking’, as well as illnesses resulting from alcohol abuse were determined. In addition, specific possible factors which influence the consumption of alcohol among students were examined. For this, demographic factors, the structure of the course studied, especially against the background of the changes in the university qualifications according to the Bologna agreement, as well as general mental pressures were examined closer. The data was collected in 2008 through an online survey of 2,348 students at three universities in Lower Saxony, namely the Technical University of Braunschweig, the University of Arts in Braunschweig and the University of Applied Sciences Braunschweig/Wolfenbüttel. Finally, an evaluation of the survey was made and an interview with 72 participants was carried out in order to establish specific pathological drinking patterns and mental disturbances. The results show a significant level of alcohol consumption. Over a period of 30 days before the survey, only 10.2% of the students involved claimed not to have drunk any alcohol , 70.8% drank without risk and 19% consumed an amount of alcohol which can be considered detrimental to health. This high level of alcohol consumption is also evident in the high frequency of ‘Binge-drinking’: 34.9% of the participants in the survey were binge-drinkers, a further 14.5% were so called ‘Heavy-users’, who consume a very large amount of alcohol(binge-drinking) on more than five days a month. According to an alcoholism screening (CAGE), 30.3% are alcohol abusers or even alcohol dependent. The results show a deviance in alcohol consumption among students as compared with their peer group in the population and with the population generally. However, an explanation of this high consumption of alcohol seems to depend less on internal conditions, for example coping with depression or anxiety, but much more on socially motivated factors, where alcohol consumption in part has a positive effect on psychological balance. Further explanations for this phenomenon will be discussed.

Financial services must monitor their transactions to prevent being used for money laundering and combat the financing of terrorism. Initially, organizations in charge of fraud regulation were only concerned about financial institutions such as banks. However, nowadays, the Fintech industry, online businesses, or platforms involving virtual assets can also be affected by similar criminal schemes. Regardless of the differences between the entities mentioned above, malicious activities affecting them share many common patterns. This dissertation's first goal is to compile and compare existing studies involving machine learning to detect and analyze suspicious transactions. The second goal is to synthesize methodologies from the last goal for tackling different use cases in an organized manner. Finally, the third goal is to assess the applicability of deep generative models for enhancing existing solutions. In the first part of the thesis, we propose an unsupervised methodology for detecting suspicious transactions applied to two case studies. One is related to transactions from a money remittance network, and the other is related to a novel payment network based on distributed ledger technologies. Anomaly detection algorithms are applied to rank user accounts based on recency, frequency, and monetary features. The results are manually validated by domain experts, confirming known scenarios and finding unexpected new cases. In the second part, we carry out an analogous analysis employing supervised methods, along with a case study where we classify Ethereum smart contracts into honeypots and non-honeypots. We take features from the source code, the transaction data, and the funds' flow characterization. The proposed classification models proved to generalize well to unseen honeypot instances and techniques and allowed us to characterize previously unknown techniques. In the third part, we analyze the challenges that tabular data brings into the domain of deep generative models, a particular type of data used to represent financial transactions in the previous two parts. We propose a new model architecture by adapting state-of-the-art methods to output multiple variables from mixed types distributions. Additionally, we extend the evaluation metrics used in the literature to the multi-output setting, and we show empirically that our approach outperforms the existing methods. Finally, in the last part, we extend the work from the third part by applying the presented models to enhance classification tasks from the second part, commonly containing a severe class imbalance. We introduce the multi-input architecture to expand models alongside our previously proposed multi-output architecture. We compare three techniques to sample from deep generative models defining a transparent and fair large-scale experimental protocol and interesting visual analysis tools. We showed that general machine learning detection and visualization techniques could help address the fraud detection domain's many challenges. In particular, deep generative models can add value to the classification task given the imbalanced nature of the fraudulent class, in exchange for implementation and time complexity. Future and promising applications for deep generative models include missing data imputation and sharing synthetic data or data generators preserving privacy constraints.