Projects per year
Introduction / mission
The chair studies data mining (DM) techniques and knowledge discovery approaches that are at the core of data science. The group is known for its contributions to the areas of predictive analytics, automation of machine learning and networked science, subgroup discovery and exceptional model mining, and similarity computations on complex data. Its research is inspired by theoretical computer science, systems development and real-world applications of (big) data-driven discovery in healthcare, banking, energy, retail, telecom, and education among others.
We develop generic approaches and specialized techniques that cover a wide range of descriptive, predictive and prescriptive analytics and work effectively with text, image, transactional, graph and time-series data in a responsible manner. E.g. we use Deep Learning methods to develop models for high dimensional heterogeneous, unstructured and evolving data and apply this models to areas such as medical imaging, genomics, anomaly detection and sentiment analysis. We further work on methods for analyzing and explaining the model’s decisions and performance and facilitate effective DM with domain expert in the loop.
We have created OpenML: an online collaborative platform for studying machine learning techniques. OpenML is used by almost 2,000 researchers, students, and practitioners world-wide, and contains around 20,000 datasets, 3,000 machine learning workflows, and 1,7 million shared experiments. It has won the Dutch Data Prize, as well as backing from Microsoft Research. It is crucial for the development of automated machine learning that is adopted by companies such as Philips.
Further information at OpenML.org
- NWO RATE-Analytics (with Tilburg University, Rabobank and Achmea) "Next generation predictive analytics for data-driven banking and insurance".
- ImpulseKYC-Analytics (with Rabobank) "Know your customer predictive analytics" project aims at developing approaches for effective DM on heterogeneous and evolving data sources with expert-in-the-loop.
- STW CAPA (with Adversitement and StudyPortals)"Context-aware predictive analytics" advanced the current state of the art in Web analytics.
- NWO Veni "Detection methods for similarity structures in time-dependent data"develops foundations for advanced time series and trajectories clustering.
- H2020 SODA (ICT-2016-1; Big Data PPP) "Scalable Oblivious Data Analytics" facilitates secure DM; together with Crypto group we develop practical approaches for DM with multi-party computation.
Research Output per year
Research output: Contribution to journal › Article › Academic › peer-review
Research output: Contribution to journal › Article › Academic
Activities per year
Supervisor: Pechenizkiy, M. (Supervisor 1) & Scheepens, R. J. (Supervisor 2)
Student thesis: Master
An exploration and evaluation of concept based interpretability methods as a measure of representation quality in neural networksAuthor: Remmits, Y., 30 Sep 2019
Supervisor: Menkovski, V. (Supervisor 1) & Stolikj, M. (External coach)
Student thesis: Master