Auflistung nach Schlagwort "Machine Learning"
1 - 10 von 108
Treffer pro Seite
Sortieroptionen
- KonferenzbeitragActivity Recognition over Temporal Distance using Supervised Learning in the Context of Dementia Diagnostics(Mensch und Computer 2022 - Tagungsband, 2022) Staab, Sergio; Bröning, Lukas; Luderschmidt, Johannes; Martin, LudgerIn the case of neurological diseases, the progression of the disease can be detected by monitoring movements and activities. Documenting such monitoring requires time-consuming work, which can hardly be covered in the context of a constantly decreasing availability of nursing staff. In cooperation with two dementia residential communities, we incrementally develop a process that supports the nursing staff by providing an approach for a semiautomated documentation. This paper presents an approach to aggregate individual activities over a care period using smartwatches in combination with supervised learning algorithms. A smartwatch offers the opportunity to integrate sensor technology into a patient’s daily routine without disturbing them, as many patients already wear watches. We are investigating promising combinations of sensor technologies and supervised learning algorithms, collecting data from the accelerometer, heart rate sensor, gyroscope, gravity and position sensor at 20 Hz and sending it to a web server. The activities are then classified multiple times using Fast Forest, Logistic Regression and Support Vector Machines over a maintenance layer. We present an activity classification prototype over time distance for automated activity recognition, which, after a number of classifications and the likelihood of these, suggests to the nurse a statement of activities over the respective time period of a nursing shift, in the form of a completed documentation. In addition, the work provides an interpretation of how the knowledge gained can be used to recognise motor skills in the course of caring for patients with neurological diseases.
- TextdokumentAIDA-Vis – Automatic Data Visualization with Human Preferences(INFORMATIK 2022, 2022) Laurito,Walter; Höllig,Jacqueline; Lachowitzer,Jonas; Thoma,Steffen; Budde,Matthias; Philipp,PatrickData visualization is a complex task that typically requires human expertise, acquired through a large number of professional working hours. The automatic generation of reasonable visualizations would be a good solution for inexperienced laypeople. However, existing approaches fall short since they are quite static and rely only on traditional supervised learning. This results in models which recommend a single visualization solely based on the dataset features. User preferences and goals are not taken into account. We propose a more flexible solution that is iteratively updated with the individual user's preferences and outputs a ranked list of visualizations for a given dataset.
- TextdokumentAnalyse von Heizungs- und Lüftungsverhalten mit Data Mining Methoden(INFORMATIK 2020, 2021) Westhäusser, Lutz; Nickel, David; Behrens, Grit; Schlender, KlausIn dem hier beschriebenen Projekt wird interdisziplinär mit Psychologen zusammen gearbeitet. Ziel der Arbeit ist es, Modelle zu entwickeln, um das Umweltverhalten von Hausbewohnern positiv zu beeinflussen und zu verstetigen. In der hier beschriebenen Arbeit werden die ersten Daten aus dem ‚Reallabor' Sennestadt genutzt, die in den Wohnungen von freiwilligen Studienteilnehmern zu ihrem Heizungs-und Lüftungsverhalten erhoben werden. Mittels Machine Learning Technologien werden diese Daten analysiert.
- ZeitschriftenartikelAnalysis of Political Debates through Newspaper Reports: Methods and Outcomes(Datenbank-Spektrum: Vol. 20, No. 2, 2020) Lapesa, Gabriella; Blessing, Andre; Blokker, Nico; Dayanik, Erenay; Haunss, Sebastian; Kuhn, Jonas; Padó, SebastianDiscourse network analysis is an aspiring development in political science which analyzes political debates in terms of bipartite actor/claim networks. It aims at understanding the structure and temporal dynamics of major political debates as instances of politicized democratic decision making. We discuss how such networks can be constructed on the basis of large collections of unstructured text, namely newspaper reports. We sketch a hybrid methodology of manual analysis by domain experts complemented by machine learning and exemplify it on the case study of the German public debate on immigration in the year 2015. The first half of our article sketches the conceptual building blocks of discourse network analysis and demonstrates its application. The second half discusses the potential of the application of NLP methods to support the creation of discourse network datasets.
- TextdokumentAnomaly Detection in Log Data using Graph Databases and Machine Learning to Defend Advanced Persistent Threats(INFORMATIK 2017, 2017) Schindler, TimoAdvanced Persistent Threats (APTs) are a main impendence in cyber security of computer networks. In 2015, a successful breach remains undetected 146 days on average, reported by [Fi16].With our work we demonstrate a feasible and fast way to analyse real world log data to detect breaches or breach attempts. By adapting well-known kill chain mechanisms and a combine of a time series database and an abstracted graph approach, it is possible to create flexible attack profiles. Using this approach, it can be demonstrated that the graph analysis successfully detects simulated attacks by analysing the log data of a simulated computer network. Considering another source for log data, the framework is capable to deliver sufficient performance for analysing real-world data in short time. By using the computing power of the graph database it is possible to identify the attacker and furthermore it is feasible to detect other affected system components. We believe to significantly reduce the detection time of breaches with this approach and react fast to new attack vectors.
- TextdokumentAnomaly Detection in Motion Timeseries using the Bosch XDK and Dynamic Time Warping(SKILL 2021, 2021) Mejía, Julián Rico; Isaías, Oscar Aguilar Aguila; Paschapur, PriyankaThis paper presents the development of an anomaly detector for robotic movements using the dynamic time warping (DTW) algorithm and its implementation in Matlab. Data was collected by mounting the Bosch Cross-Domain Development Kit (XDK) sensor on a collaborative robot arm (Cobot), aiming at industrial applications in need for motion anomaly detection during repetitive tasks. The paper discusses practical issues like parameter tuning as well as algorithmic variants such as decoupling accelerometer and gyroscope data.
- TextdokumentAnwendung von Machine Learning bei der datengetriebenen Prozessanalyse - Eine State-of-the-Art Literaturanalyse(INFORMATIK 2022, 2022) Welz,Laslo; Beckmann,HelmutDie Disziplin Process Mining im Anwendungsbereich der datengetriebenen Prozessanalyse ermöglich die Abbildung realer Geschäftsprozesse durch die Extrahierung von Daten aus Eventlogs von Informationssystemen. Konventionelles Process Mining kann mit Ansätzen aus dem Bereich Machine Learning ergänzt werden, um die Prozessanalysen zu verbessern. Anhand einer Literaturanalyse untersucht diese Forschungsarbeit die Anwendung von Machine Learning bei Process Mining. Die Ergebnisse aus einer Stichprobe von 34 Publikationen zeigen, dass in den beiden Process Mining Bereichen „Discovery“ und „Enhancement“ die meisten Machine Learning-Methoden angewendet werden. Insbesondere ist die Anwendung von Entscheidungsbäumen und Neuronalen Netzen weit verbreitet.
- KonferenzbeitragApproach to Synthetic Data Generation for Imbalanced Multi-class Problems with Heterogeneous Groups(BTW 2023, 2023) Treder-Tschechlov, Dennis; Reimann, Peter; Schwarz, Holger; Mitschang, BernhardTo benchmark novel classification algorithms, these algorithms should be evaluated on data with characteristics that also appear in real-world use cases. Important data characteristics that often lead to challenges for classification approaches are multi-class imbalance and heterogeneous groups. Real-world data that comprise these characteristics are usually not publicly available, e. g., because they constitute sensible patient information or due to privacy concerns. Further, the manifestations of the characteristics cannot be controlled specifically on real-world data. A more rigorous approach is to synthetically generate data such that different manifestations of the characteristics can be controlled. However, existing data generators are not able to generate data that feature both data characteristics, i. e., multi-class imbalance and heterogeneous groups. In this paper, we propose an approach that fills this gap as it allows to synthetically generate data that exhibit both characteristics. In particular, we make use of a taxonomy model that organizes real-world entities in domain-specific heterogeneous groups to generate data reflecting the characteristics of these groups. In addition, we incorporate probability distributions to reflect the imbalances of multiple classes and groups from real-world use cases. Our approach is applicable in different domains, as taxonomies are the simplest form of knowledge models and thus are available in many domains. The evaluation shows that our approach can generate data that feature the data characteristics multi-class imbalance and heterogeneous groups and that it allows to control different manifestations of these characteristics.
- KonferenzbeitragAssessing the performance of Neural Networks in Recognizing Manual Labor Actions in a Production Environment(INFORMATIK 2023 - Designing Futures: Zukünfte gestalten, 2023) Höfinghoff, Maximilian; Buschermöhle, Ralf; Korn, Goy-Hinrich; Schumacher, Marcel; Seipolt, ArneAction recognition technology has gained significant traction in recent years. This paper focuses on evaluating neural network architectures for action recognition in the production industry. By utilizing datasets tailored for production or assembly tasks, various architectures are assessed for their accuracy and performance. The findings of this study provide some insights and guidance for researchers and practitioners to select an appropriate architecture or pretrained models for action recognition in the production industry.
- ZeitschriftenartikelAutomated Traceability Link Recovery Between Requirements and Source Code(Softwaretechnik-Trends Band 44, Heft 2, 2024) Hey, TobiasEfficient development, maintenance, and management of software systems rely heavily on understanding the relationships between the various software artifacts. Manual creation and maintenance of traceability information between these artifacts incur high costs due to the required human expertise. Often this results in a deficiency in available traceability information that hampers the efficiency of the software projects. The FTLR approach presented in this dissertation aims to enhance automatic traceability link recovery between requirements and source code by leveraging a fine-grained semantic similarity comparison using pre trained word embeddings and Word Movers Distance. FTLR achieves significant improvements in traceabil ity connection identification by employing the fine grained mapping with subsequent majority vote-based aggregation. Additionally, it employs a novel approach for filtering irrelevant parts of requirements using a large language model-based classifier called NoRBERT, which achieves promising results on unseen projects. Furthermore, this dissertation explores the integration of bimodal large language models into FTLR but finds no significant performance increase over word embeddings. A comparative analysis showed that FTLR outperforms existing approaches in mean average precision and F1-scores, especially on projects with object-oriented source code. However, challenges, remain to fully automate traceability link recovery, particularly in large-scale projects.