ADHS ML

In the project ADHS ML machine learning processes are used to predict hyperactivity of children with the help of accelerometers.

Up to now it is common to investigate hyperactivity of children via magnetic resonance imaging (MRI). The recorded images are used for gaining a classifier to make a unique diagnosis. Although the classification rates are very promising the method implies disadvantages concerning temporal and financial costs. In addition, the survey of hyperactivity cannot take place in every-day surroundings. To capture this, there are many common and low cost alternatives such as accelerometers, in contrary to the expensive and resource intense MRI experiments. Especially, accelerometers play a major role in the classification of physical activities of people, because they can largely automatise the observation of patients in the context of health research.

Objectives

1. Identification of hyperactivity regarding children

2. Survey of hyperactivity in everyday-surroundings

3. Using intelligent methods of machine learning

Method

The project aims at investigating the identification of hyperactivity regarding children. The measurements of children with accelerometers in different scenarios serve as data sets (e.g., playing cards). We are using intelligent methods of machine learning, especially support-vector- machines. In contrary to Johnson et al. (2012), our approach offers a low cost and easy possibility to detect hyperactivity that does not limit the child’s degree of freedom.

 

Funding: Loewe

Cooperation: IDea ADHS

Duration: 2013 - 2014

Project Manager: Caterina Gawrilow

Contact: Ulf Brefeld

Links: IDeA ADHS

 

Literature:

- U. Brefeld, C. Büscher, and T. Scheffer: Multi-view Discriminative Sequential Learning. Proceedings of the European Conference on Machine Learning, 2005.

- U. Brefeld and T. Scheffer: Semi-supervised Learning for Structured Output Variables. Proceedings of the International Conference on Machine Learning, 2006.

- E. R. Fernandes and U. Brefeld: Learning from Partially Annotated Sequences. Proceedings of the European Conference on Machine Learning. 2011.