Using A Penalized Maximum Likelihood Model For Feature Selection
Amir Jalalirad, Tjalling Tjalkens

Feature selection and learning through selected features are the two steps that are generally taken in classification applications. Commonly, each of these tasks are dealt with separately. In this paper, we introduce a method that optimally combines feature selection and learning through feature-based models. Our proposed method implicitly removes redundant and irrelevant features as it searches through a comprehensive class of models and picks the penalized maximum likelihood model. The method is proved to be efficient in terms of the reduction of the calculation complexity and the accuracy in the classification of artificial and real data.