University of Konstanz
Graduiertenkolleg / PhD Program
Computer and Information Science

Iris Adä

Associated Doctoral Student in the PhD program since September 15, 2009.


Prof. Dr. Michael Berthold

organisational data

Room: Z 815
Tel.: +49 (0)7531 / 88-4719
E-mail: adae "at"
Other Resources: Personal Page at Workgroup

project description

Event detection in time series data A long time ago the collection of data was a very rare process. At a total maximum every minute a new value was measured. But with the exponentialspeed of technical development, data points can now even be achieved in real time; Online information can be summarized as sensor data. In the area of analyzing and monitoring this information, a subfield of data mining, which is called stream mining, interesting additional requirements have to be taken into account. The data has to be processed online, which means it is never possible to consult all the data, or only all previous data. Secondly the algorithm must be able to predict a new data point at any time.
One very interesting challenge in data stream mining is called event detection. An event is defined as anything irregular in the data. This could simply be an outlier or an incorrectly classified pattern. But with more sophistication, an event can also be caused by a change or drift in the data. My research mainly focuses on these two aspects: Detecting an event in an online data stream and evaluating which type of event occurred.
Our first approach yields in monitoring the data stream using an idea adapted from meta learning for data streams. To monitor the stream different base learners are used to model different parts of the data stream. A new base learner is started with every data point, an old learner is stopped, and the achieved model is saved. By comparing these base learners an event can be detected if there is a big difference or distance between the learners. The approach was successfully applied with Gaussian Mixture Models as base learners to model a multidimensional real-valued data stream. For the model comparison an approximation of the Kullback-Leibler-Divergence was used.
Another recent development is the EVE-Framework¹ . As previously mentioned, various methods for event detection have been proposed for different types of events. However, a lot of them make the same prior assumption, following the core idea behind EVE. EVE is a more general framework for event detection. The framework enables generic types of time slots and streaming progress to be incorporated through time. It allows measures of similarity to included between those slots, either based directly on the data, or on an abstraction, e.g. a model built on the data. A large number of existing algorithms fit nicely into this framework by choosing appropriate window combinations, progress mechanisms, and similarity functions.

¹Adä, Iris, Berthold, Michael R., Unifying Change - Towards a Framework for De tecting the Unexpected. In: Data Mining Workshops (ICDMW), 2011 IEEE 11th International Conference on Data Mining


The following list of publications covers only those, which are or were published during participation at the Graduiertenkolleg / PhD program.

Articles in Journals


Conference Papers

  • Fillbrunn, A., Adä, I., Gabriel, T., Berthold, M., Ensembles and PMML in KNIME, Proceedings of the PMML Workshop, 2013. abstract
  • Adä, I., Berthold, M., Detecting Events in Molecular Dynamics Simulations, Proceedings of the 12th International Symposium on Intelligent Data Analysis (IDA 2013), Allan Tucker and Frank Höppner and Arno Siebes and Stephen Swift (ed.) , Vol. 8207, pp. 44-55, London, UK, 2013. abstract
  • Adä, I., Berthold, M., Unifying Change - Towards a Framework for Detecting the Unexpected, IEEE 11th International Conference on Data Mining (IEEE11), Data Mining Workshops (ICDMW), 2011, IEEE Press. abstract

curriculum vitae

since September 2009 Associate member of the graduate school "Explorative Analysis and Visualization of large Information Spaces"
since July 2009 Research assistant at the University of Konstanz, Department of Computer and Information Science
October 2004 - May 2009 Study of Computer Science and Maths at the University of Konstanz