TY  - CHAP
ID  - moewes_simple_2008
T1  - Simple Machine Learning Approaches to Safety-Related Systems
A1  - Moewes, Christian
A1  - Otte, Clemens
A1  - Kruse, Rudolf
ED  - De, Rajat Kumar
ED  - Mandal, Deba Prasad
ED  - Ghosh, Ashish
TI  - Machine Interpretation of Patterns: Image Analysis and Data Mining
T3  - Statistical Science and Interdisciplinary Research
Y1  - 2010
VL  - 11
SP  - 231
EP  - 249
U2  - Chapter: 12
PB  - World Scientific Publishing Co. Inc.
AD  - Hackensack, NJ, USA
SN  - 978-981-4299-18-3
UR  - http://dx.doi.org/10.1142/9789814299190_0012
M2  - doi: 10.1142/9789814299190_0012
KW  - Multiple-Instance Learning
KW  - Safety-Related Systems
KW  - Support Vector Machine
N2  - The principles of machine learning become gradually more interesting to safety-related applications. This paper introduces a preprocessing method for such kind of applications. Here we concentrate on scenarios with highly unbalance misclassification costs. Therefore we briefly introduce a variation of multiple-instance learning (MIL) and recall soft margin hyperplane classifiers. According to this classifier we present a training set selection method for multidimensional problems that combines the idea of support vector pruning with pattern weighting. The proposed method guarantees both high performance and interpretable decisions. The main advantage of our proposed filter method is the obtained training set that rather prefers quasilinear classifiers during model selection. We conclude with potential extensions for future research in this domain.
ER  -