Objective feature selection is an important component in the machine learning framework, which has addressed problems like computational burden increasing and unnecessary high-dimensional representations. Most of feature selection techniques only perform individual feature evaluations and ignore the structural relationships between features of the same nature, causing relations to break and harming the algorithm performance. In this paper a feature group selection technique is proposed with the aim of objectively identify the relevance that a feature group carries out in a classification task. The proposed method uses Multiple Kernel Learning with a penalization rule based on the ℓ1 -norm and a Support Vector Machine as base learner. Performance evaluation is carried out using two binarized configurations of the freely available MFEAT dataset. It provides six different feature groups allowing to develop multiple feature group analysis. The experimental results show that the implemented methodology is stable in the identification of the relevance of each feature group during all experiments, what allows to outperform the classification accuracy of state-of-the-art methods.
CITATION STYLE
Areiza-Laverde, H. J., Díaz, G. M., & Castro-Ospina, A. E. (2018). Feature group selection using mkl penalized with ℓ1 -norm and svm as base learner. In Communications in Computer and Information Science (Vol. 915, pp. 136–147). Springer Verlag. https://doi.org/10.1007/978-3-030-00350-0_12
Mendeley helps you to discover research relevant for your work.