Print
Univis
Search
 
FAU-Logo
Techn. Fakultät Willkommen am Institut für Informatik FAU-Logo

Feature Extraction and Selection

Optimal Feature Selection for Nonlinear Data using Branch-and-Bound in Kernel Space
  • Branch-and-bound (B&B) feature selection finds optimal feature subsets without performing an exhaustive search. The classification accuracy achievable with optimal B&B feature subsets, however, is often inferior compared to the accuracy achievable with other algorithms that guarantee optimality.

    We argue this is due to the currently used criterion functions that define the optimal feature subset but may not conceive inherent nonlinear data structures. Therefore, we propose B&B feature selection in Reproducing Kernel Hilbert Space (B&B-RKHS). B&B-RKHS employs two currently used criterion functions (Bhattacharyya distance, Kullback-Leibler divergence) and one new criterion function (mean class distance), however, all computed in RKHS. This enables B&B-RKHS to conceive inherent nonlinear data structures.

    B&B-RKHS was experimentally compared to the popular wrapper approach for feature selection, which requires an exhaustive search to guarantee optimality. The classification accuracy achieved with both methods was comparable. However, runtime of B&B-RKHS was superior using the two existing criterion functions and even completely out of reach using the new criterion function (about 1,500 times faster on average).

    Opens internal link in current windowImplementation available in the ECST

    Journal Articles
    Optimal feature selection for nonlinear data using branch-and-bound in kernel space
    Pattern Recognition Letters, vol. 68, pp. 56-62, 2015 (BiBTeX, Who cited this?)
Generic Features for Biosignal Classification
  • The recent progress in sensor technology enables wearable support and monitoring tools for sports applications. Defining a suitable representation of the measured data is a current challenge in data analysis with pattern recognition methods. We therefore compiled a generic feature set as data representation and proved its applicability in two biosignal analysis applications. Physiological electrocardiogram (ECG) data were used to classify activity levels with a classification rate of 88.8% and kinematic inertial sensor golf putt data were used to classify golf experience levels with a classification rate of 90.2%. The proposed feature set was integrated on a freely available software package that facilitates rapid prototyping of classification systems from a sports expert perspective. 

    Articles in Conference Proceedings
    Sportinformatik 2012 : 9. Symposium der Sektion Sportinformatik der Deutschen Vereinigung für Sportwissenschaft (9. Symposium der Sektion Sportinformatik der Deutschen Vereinigung für Sportwissenschaft), Konstanz, Germany, 12.-14.09.2012, pp. 162-168, 2012 (BiBTeX, Who cited this?)
Classification of Kinematic Golf Putt Data with Emphasis on Feature Selection
  • The complex movement sequences of golf require supporting tools for players and coaches alike. We developed a system that classifies the experience level and trained it with data from an inertial sensor on the club head. Based on 315 golf putts from eleven subjects the system differentiated between experienced and unexperienced players with a classification rate of 86.1%. To improve the classification system and obtain discriminant features we additionally integrated a feature selection step. We compared different selection approaches and concluded that a leave-subject-out feature selection was the appropriate approach to predict the true performance of a live system. The selected features can be fed back to coaches and help them to guide players to a better putting technique.

    Articles in Conference Proceedings
    Jensen, Ulf; Dassler, Frank; Eskofier, Björn
    Proceedings of the 21st International Conference on Pattern Recognition (21st International Conference on Pattern Recognition (ICPR 2012)), Tsukuba Science City, Japan, November 11-15, pp. 1735-1738, 2012 (BiBTeX, Who cited this?)