This is an archived version of the Machine Learning Lab website for documentation purposes. The host is in no way affiliated with the University of Freiburg, its Faculty of Engineering or any of its members.

Classification of Nuclear Contamination Measurements


In this joint project with the Federal Research Center for Nutrition we were faced with the task of classifying data in terms of trustworthyness. The data were taken from a database of nuclear contamination of nutrition which is used to control the development of nuclear contamination in Germany. Due to the large amount of data, an automatic classification tool was developed. This field of research is also known as Novelty detection, Outlier Detection, or Single-class classification.

The data exhibited a lot of missing values and measurements that were only determined down to a certain lower bound. Thus, we had to integrate techniques of missing value estimation like Data Augmentation and Survival analysis. Furthermore, the actually continuously distributed data showed significant artefacts of discretization.

Our approach is based on a robust density estimate of the data. We use Gaussian mixture densities as model and a randomized learning algorithm based on Markov Chain Monte Carlo techniques. The appropriate size of the model, as well, is determined using the Monte Carlo approach. The model and the learning algorithm were implemented in a tool that can easily be used for outlier classification. It is adapted to the needs of our partners.


The project was conducted in the years 2000 and 2001 at the University of Karlsruhe under the direction of Prof. Dr. Wolfram Menzel.


The Federal Research Center for Nutrition is concerned with studies on healthy nutrition with special emphasis on vegetables and fruit. The field of work refers to post harvest behavior of agricultural products, health promoting effects and hygenic quality assurance as well as consumer attitudes concerning food and nutrition.


Researchers working on this project:

  • Prof. Dr. Wolfram Menzel
  • Prof. Dr. Martin Riedmiller
  • Martin Lauer



For more information on this research project, please contact Martin Riedmiller.